Sage Journals: Discover world-class research

Abstract

This study introduced a framework that facilitates bridge diagnostic and prognostic evaluations and related visualizations, leveraging building information modeling (BIM) techniques and data-driven structural health monitoring. Diagnostic and prognostic algorithms are proposed and able to provide intuitive and accessible structural insights via three-dimensional visualizations supported by the BIM techniques. Through the data communication channel established, virtual model information was successfully embedded into the bridge health evaluations, emphasizing the contribution of BIM techniques for data analysis. The proposed algorithms were applied to a case study of a reinforced concrete bridge. Strain distribution along the bridge deck was mapped in the 3D virtual model, which was able to precisely identify the critical location for potential damage. The residual fatigue life of different parts of the bridge deck was estimated and visualized by considering the main effects of traffic loading, exhibiting good alignments with the diagnostic results. The proposed framework is able to provide understandable and actionable structural information for stakeholders, facilitating the timely identification of structural issues and long-term planning of maintenance schedules.

Keywords

Structural health monitoring building information modeling data visualization bridge diagnosis bridge prognosis bridge management

Introduction

Structural health monitoring (SHM) systems, employing an array of sensors such as strain gauges (SGs), accelerometers, and thermometers, offer continuous, real-time, and objective monitoring of the behavior of infrastructures. With the advancement of sensor technologies, SHM systems have gained increasing recognition in the field of bridge health monitoring and management, where diagnostic and prognostic evaluations constitute two important components of bridge SHM data analysis and interpretation, working collaboratively to ensure structural integrity and longevity. In detail, diagnosis involves the interpretation of monitoring data to detect, locate, and characterize any damage or deterioration in bridge structures. Prognosis, extending beyond diagnosis, involves predicting the future condition of the bridge based on current diagnostic data and the historical performance of the structure. A number of bridges, such as the Tsing Ma Bridge,¹ the Golden Gate Bridge,² and the Akashi-Kaikyo Bridge,³ have adopted SHM systems for efficient and labor-saving health monitoring.

In the field of current bridge diagnostic and prognostic evaluations, methodologies can be broadly divided into data-driven and model-based algorithms.^4–6 Data-driven algorithms are increasingly favored for assessing aging bridges, especially in cases where original design data may be lacking. This preference is due to the fact that data-driven algorithms do not require a detailed prior understanding of the physical properties or behavior of bridge structures. Furthermore, these algorithms leverage statistical^7,8 and machine learning techniques^9,10 to analyze large amounts of sensor data. They have been effectively implemented in the domains of bridge damage detection, condition assessment, and predictive maintenance.^11–13 For instance, Pan et al.¹⁴ employed main feature extraction techniques, including the Wavelet transform, the Hilbert-Huang transform, and the Teager-Huang transform, to extract time-frequency characteristics from vibration data. The resulting features, such as time-frequency representation points, were adopted to identify the presence of damage on cable-stayed bridges. Similarly, Wang et al.¹⁵ utilized sparse Bayesian learning and vibration data to identify damage-sensitive frequency bands for detecting damage. To predict the lifespan of bridge structures, Wang et al.¹⁶ proposed a bridge dynamic deformation index and a temperature-time-deformation model to assess and forecast long-term deformation. However, it should be noted that the output from current data-driven practices based on on-site collected sensor data requires further interpretation, difficult to directly translate into actions for bridge maintenance and health management, especially for stakeholders from different backgrounds. Moreover, these diagnostic algorithms primarily analyze vibrational data to evaluate the overall health of a bridge, thus offering a limited role in locating damage.

On the other hand, the massive volume and complexity of SHM data present significant challenges in terms of its interpretation and management. Specifically, an SHM system can generate gigabytes of data daily, all of which require proper storage, processing, and analysis.¹⁷ To enhance the interpretation and management of SHM data for bridge maintenance and health management, various visualization and management techniques have been employed.^18–22

Building information modeling (BIM) techniques, which originated from building lifecycle management, have achieved significant accomplishments in bridge design and construction phases²³ for effective data management and visualization. Recently, their application in bridge operation and maintenance (O&M) phases is gaining increasing attention. Traditionally, research has focused on utilizing BIM as an information repository and developing models to support bridge management systems (BMSs) through visualizations of on-site inspection data.²⁴ For instance, McGuire et al.²⁵ created a self-defined damage cube within the BIM environment to document and visualize the geometries and locations of bridge damage due to corrosion, based on on-site inspection data. Similarly, Marzouk et al.²⁶ employed BIM techniques to develop a detailed database and inspection sheets, resulting in an efficient and accurate BMS that visualized inspection results through a link in Navisworks Manage software. Further integrating BIM with BMSs, Jeong et al.²⁷ proposed an information repository framework by integrating the extended OpenBrIM standard with a NoSQL database, allowing the creation of finite element models from BIM-documented geometrical information. They also introduced methodologies for modeling sensor networks, enhancing the integration of BIM with model-based SHM.

In parallel, the prosperity of data-driven SHM has directed research toward creating BIM models for documenting and visualizing complex SHM data. For example, Boddupalli et al.²⁸ visualized sensor schedule table in Revit, documenting the links to files containing sensor information and the visualizations of both bridge static and dynamic analysis results. Deng et al.²⁹ developed a front-end visualization platform using Revit application programming interface to visualize and share safety warnings of bridge health status by setting thresholds for strains collected from on-site sensors. By utilizing Internet of Things (IoT) sensors and BIM techniques, Sakr and Sadhu³⁰ enabled real-time visualizations of diagnostic information for the monitored structure, incorporating both time and frequency domain analyses based on sensor data collected from laboratory experiments.

Despite these achievements, the application of BIM techniques to the O&M phases of the bridge lifecycle is still under development, particularly in their integration with data-driven SHM. Specifically, BIM models are still limited in documenting inspection and evaluation results, offering general maintenance advice and limited structural information. Additionally, visualizations based on the integration of BIM techniques and data-driven SHM are primarily achieved through back-end software, with BIM platforms merely storing links to these visualizations, lacking direct display within a 3D BIM model. Moreover, current studies mainly focus on analyzing and visualizing global health characteristics of the monitored bridge, such as natural frequencies, offering limited value in the monitoring of local degradations.

This study introduces a framework that integrates BIM techniques with data-driven SHM, enhancing the visualization of evaluation results within a 3D BIM model. Different from previous research, information documented in the BIM model is actively involved in the evaluation process, particularly the diagnostic evaluations, where a diagnostic algorithm is developed to process strain data collected on-site and is able to locate and visualize potential damage utilizing the geometrical information documented in the BIM model. In parallel, a prognostic algorithm grounded in fatigue theories was formulated to estimate the residual fatigue life of bridge components in terms of years. To improve local health monitoring of bridges, this study divides the monitored structure into finer partitions within the BIM environment, enabling the detailed documentation and visualization of the structural properties of bridge components and information of sensors. The practical application of these methodologies was demonstrated using a two-span reinforced concrete culvert bridge to validate the effectiveness of the framework and developed algorithms, where scan-to-BIM method was adopted to construct the case study bridge in the BIM environment.

Methodology

This section begins with an introduction to the proposed framework, followed by a detailed explanation of the proposed diagnostic and prognostic algorithms which are based on field strain measurements. Then, the data communication between the data-driven SHM process and the BIM environment and its visualizations are demonstrated.

Framework introduction

The primary objective of the framework proposed in this study is to provide accessible interpretations of SHM data for better bridge health management. Four phases of the SHM data processing are included in this framework, encompassing data collection at the bridge site (Phase I), data storage with databases (Phase II), data analysis with analytical tools and algorithms (Phase III), and data visualization with BIM techniques (Phase IV). Effective data communications are established among the SHM data processing phases, facilitating interactive visualization of the diagnostic and prognostic data, and therefore offering valuable and understandable structural insights for bridge integrity and longevity.

Figure 1 demonstrates the workflow of the proposed framework, which is composed of two parts: the data-driven SHM process and the BIM environment. In detail, the data-driven SHM process employs MATLAB as the analytical tool to perform diagnostic and prognostic evaluations, dedicated to extracting structural information relevant to bridge health. In parallel, the BIM environment leverages Revit-Dynamo as its foundational platform for the visualization of SHM data. Meanwhile, the MySQL database was adopted to store both the BIM model information and the SHM data due to its high compatibility with both the Revit-Dynamo environment and MATLAB, thus functioning as a crucial link between the BIM environment and the data-driven SHM process.

Figure 1.

The workflow and data flow of the proposed framework.

The framework intends to provide automatic data flow among the four phases of SHM data. As shown in Figure 1, on-site collected raw data, including time-series data and geometrical data, is automatically transferred and stored in the MySQL database.³¹ Sequentially, the geometrical data is queried using Dynamo for automatic model construction within Revit,³² while the time-series data is queried and processed with MATLAB for diagnostic and prognostic evaluations. To actively involve the BIM model in the diagnostic and prognostic evaluations, associated model information is transferred to the analytical tool through the links built by the database. The outcomes of these evaluations are then synchronized with the BIM model information and re-routed into Revit-Dynamo for visualization. Consequently, diagnostic and prognostic insights are firmly connected and visualized with the BIM model.

Diagnostic and prognostic algorithms

Diagnostic algorithm

In this study, the algorithms developed for bridge diagnosis and prognosis are specifically tailored to the identification and visualization of critical areas on bridge deck, enhancing the clarity and accessibility of diagnostic and prognostic data. Strain data was selected for diagnostic analysis in this study due to its sensitivity to local structural damage and changes.^33,34 The spatial distribution of strain across the bridge deck yields critical insights into both structural integrity and traffic dynamics, identifying the most deformed area on the deck and providing traffic patterns. To cost-effectively map the spatial strain distribution on the deck, the values collected by the SGs on-site are used to extrapolate the strains of adjacent areas according to specific rules detailed below.

It is assumed that when traffic load acts at the location of the SG, the strain distribution in the affected area of the load reaches its maximum at the location of the SG and decreases progressively with distance from the gauge following predefined strain mapping rules. The influence zone of each SG is defined as a circular area, centered at the location of the gauge. The strain value at a given location on the deck is determined as

Strain = R \times S

(1)

where $Strain$ denotes the strain value at the given location. $S$ is the strain captured by the adjacent SG, and $R$ represents the contribution of the SG. In this study, two strain mapping rules, that is, linear and normal distribution, are considered for the calculation of $R$ . In detail, by assuming a negative linear relationship between the strain value and its distance from the gauge, the linear strain contribution coefficient (SCC) $R_{l}$ can be determined as

R_{l} = k \times d + b

(2)

where $d$ represents the distance between the given location and the SG. $k$ and $b$ are parameters determined by the size of the influence zone.

On the other hand, by assuming a more gradual decrease in strain values from the center of the influence zone and consequently, more conservative results, a normal distribution rule is also considered to calculate the corresponding SCC $R_{n}$ as

R_{n} = e^{- \frac{d^{2}}{2 \times σ^{2}}}

(3)

Similarly, the standard deviation $σ$ is also related to the size of the influence zone.

Prognostic algorithm

In this study, the prediction of residual fatigue life is performed in two steps using field strain measurements, employing the classical approach^35,36 of combining the S − N curve with the cumulative damage law.³⁷ A schematic flowchart of the fatigue analysis is presented in Figure 2. In detail, residual fatigue cycles are derived from the strain-time history through a sequence of processes, involving conversion to stress-time history, rain-flow counting, S − N curve analysis, and calculations based on Miner’s rule.³⁷ Subsequently, the obtained results are converted into residual fatigue years, utilizing an appropriate assumption that correlates fatigue cycle consumption with traffic volume growth.

Figure 2.

Schematic flowchart of residual life estimation.

In detail, on-site collected strain data is first converted to stress data by employing Young’s modulus relationship.³⁸ Following this conversion, the stress data is processed with the rain-flow counting algorithm, resulting in the generation of a stress-spectrum histogram.³⁹ Subsequently, the equivalent effective stress is calculated correspondingly as

S_{eff} = {(\sum n_{i} S_{n_{i}}^{3})}^{1 / 3}

(4)

where $n_{i}$ is the times of the ith stress amplitude ( $S_{n_{i}}$ ) that appeared in the stress histogram calculated by the rain-flow counting method.

By employing Miner’s rule, which allows for the quantification of cumulative fatigue damage by correlating the stress cycles to their respective fatigue life contributions, the effective fatigue cycles that the bridge has sustained can be expressed as

\frac{N_{eff}}{N} = \sum \frac{n_{i}}{N_{i}}

(5)

where $N_{eff}$ is the effective cycles of the structure, $N$ is the overall total number of cycles determined from the equivalent effective stress using S − N curves, and $N_{i}$ is the total number of cycles to fatigue failure.

For concrete bridge deck reinforced with steel rebars, which can be considered as a combination of several reinforced concrete beams, $N_{i}$ in Equation (5) can be determined using the corresponding stress range $S_{i}$ , which can be interpreted using the proposed S − N curve formula on reinforced concrete structures⁴⁰

S_{i} = 1330 (1 - 0.13 \log N_{i})

(6)

In this case, each stress range $S_{i}$ will have its corresponding total number of fatigue cycles $N_{i}$ and the effective stress range $S_{eff}$ will then produce $N$ in Equation (5).

It is understood that vehicles with similar weight and speed running through the bridge would generate similar responses to the bridge. Consequently, after analyzing the data collected from the same SG with the rain-flow counting algorithm, the number of fatigue cycles consumed is expected to show minimal variation. On the assumption that the traffic component and the speed limit on the bridge are constant from its opening, the number of fatigue cycles consumed is therefore linearly related to the traffic volume. Given that the focus of this study is the bridges primarily subjected to traffic-induced damage, other factors that may contribute to the fatigue damage, such as the environmental effects, corrosion, and physical impacts, are neglected in this analysis. With these assumptions, this study modifies traditional life estimation methods, which typically generate fatigue cycles only, to the direct calculation of structural residual life in terms of year.

In detail, the linear relationship between the traffic volume and fatigue cycles consumed in the same year is assumed as

\frac{N_{1}}{V_{1}} = \frac{N_{2}}{V_{2}} = \frac{N_{3}}{V_{3}} = \dots

(7)

where $N_{1}$ , $N_{2}$ , $N_{3}$ , … represent the number of cycles consumed each year starting from construction, and $V_{1}$ , $V_{2}$ , $V_{3}$ , … represent the traffic volumes in the corresponding year.

The total consumed number of cycles is the sum of the number of cycles consumed each year from construction to date, which can be illustrated as the sum of a geometric sequence by assuming a constant growth rate of traffic volume k per year as

N_{tp} = \frac{V_{p}}{V_{0}} N_{eff} \times \frac{365}{t} \times \frac{(1 - {(1 + k)}^{n})}{(1 - (1 + k))}

(8)

where $N_{tp}$ is the total consumed number of cycles and $N_{eff}$ is the equivalent effective cycles as determined previously using Equation (5); $V_{p}$ and $V_{0}$ represent the then and current traffic volumes, respectively; $\frac{365}{t}$ is used to convert daily traffic volume to annual; $t$ is the sampling time of the SGs in terms of day; $n$ is the number of years from construction.

The residual number of cycles of the bridge, $N_{tf}$ , can then be determined by subtracting $N_{tp}$ from the total life $N$ . The residual life of the bridge $N_{tf}$ in terms of year can finally be determined using the same principle as Equation (8) as

N_{tf} = \frac{V_{f}}{V_{0}} N_{eff} \times \frac{365}{t} \times \frac{(1 - {(1 + k)}^{n})}{(1 - (1 + k))}

(9)

where the only unknown variable is the number of residual years, n, which can be inversely calculated.

Data communication and visualization

To establish effective and precise connections between the Revit model and the diagnostic and prognostic analyses, and then achieve data visualizations within the BIM environment, it is crucial that the virtual model in Revit aligns with the physical structure in the real world. This alignment encompasses not only the primary bridge structures, such as the deck and piers but also the SHM system installed on the bridge. Sensors belonging to the SHM system should be accurately placed in the virtual model, mirroring their real-world layout. In accordance with the hierarchy-based paradigm employed by Revit,⁴¹ the process of sensor modeling should proceed through a sequence of steps: selecting the appropriate category, creating a family, developing a type, and placing instances. Furthermore, it is essential to define instance parameters that reflect the functions of the sensors, thereby enabling the incorporation of data passed from the SHM system. Figure 3 presents an example of modeling SGs. A new family is created within the Data Devices category, incorporating a new type with a geometry of $150$ × $250$ mm. Instance parameters are created to store SHM information, such as sensor name, sensor type, residual life calculated as well as strain value collected. Instance parameters start_time and end_time are created to query data within a certain time frame from the MySQL database. The newly created instances of the SG type are precisely positioned in the Revit virtual model, effectively representing the corresponding real-world SGs installed on the bridge. Another critical aspect of model alignment is to keep the naming of sensors consistent throughout the workflow of the framework. This consistency is essential to ensure that sensor data is accurately and automatically mapped to its corresponding instance in the model.

Figure 3.

Hierarchy of the sensor family.

Based on the model alignments, data communication within the proposed framework is achieved through the established communication channels and a bidirectional mapping. Communication channels are established through Dynamo and MATLAB scripts, while MySQL database, gathering data from both MATLAB and Revit-Dynamo, is a crucial link for the mapping process. Raw data collected on-site is periodically transferred to the database, and the diagnostic and prognostic procedures are carried out on demand.

Figure 4 illustrates the detailed interactions between MATLAB, MySQL, and Revit-Dynamo in the proposed framework with an example of the database structure utilized. A schema (database) called Bridge is created with tables to hold data from different sources and for different purposes. For example, strain data collected on-site is periodically saved in table Raw data with timestamps as the primary key, and the strains from different SGs are stored in the corresponding columns. To pass the strain data to the BIM model, the element IDs and names of the virtual entities (VEs) of SGs are extracted and stored in table BIM model by adopting Extraction codes (a) and Insertion codes (b) as shown in Figure 5. The connection between the database and the BIM model is established with Connection codes (c) in Figure 5, while Query Codes (d) and Setting Codes (e) are adopted to query the evaluation results and update the corresponding parameters in the Revit model.

Figure 4.

Data communication between MySQL, MATLAB, and the BIM model.

Figure 5.

Examples of dynamo codes utilized for the communication between MySQL and the BIM model.

The database and MATLAB are connected using Connection codes (f) in Figure 6. MATLAB is configured with Open Database Connectivity (ODBC) drivers to ensure a robust and reliable connection. With the established tables Raw data and BIM model, the mapping process begins by querying both tables using Query Codes (g) in Figure 6, and the mapping results are returned to the Bridge schema as table Strain value through Insertion Codes (h). The mapping of the strain at a specific timestamp is detailed in Figure 7. Based on the consistency of the SG names, the strain values in table Raw data are aligned with the corresponding element IDs in table BIM model and inserted into the table Strain value. Dynamo then reads the Strain value table to update the Revit model, where the element IDs are read to locate the corresponding virtual SGs, and the strain value parameters of those VEs are populated with the corresponding strain values.

Figure 6.

Examples of MATLAB codes utilized for the communication between MATLAB and MySQL.

Figure 7.

Strain data mapping example.

To facilitate the visualization of bridge local health, this study proposed to divide the monitored structure into finer partitions in the Revit model employing the Create Parts tab in the toolbar. Similar to the modeling of SGs, each partition is modeled with parameters to accommodate the required structural properties, allowing the visualization of structural properties in finer units. Specifically, to visualize the distribution of strains on the bridge deck, the distances between the virtual SGs and the finer partitions are calculated and then passed to the analytical tool, involved in the diagnostic analysis. Sequentially, the strains obtained are assigned to the finer partitions as parameters for visualization following the mapping process described. In this visualization scheme, varying colors are adopted to represent different strain intensities on the deck. Similarly, in the prognostic visualization, estimated residual fatigue years are mapped back to the BIM model as instance parameters of virtual SGs, with color variations indicating the likelihood of fatigue failure.

Case study

The Governor Macquarie Drive (GMD) bridge, situated in the suburb of New South Wales (NSW), Australia, is presented as a case study. Built approximately 30 years ago, this bridge consists of concrete culverts with two spans and three shear walls. In February 2018, the GMD bridge was extended as shown in Figure 8. Figure 8(a) provides a 3D view of the GMD bridge, including its detailed dimensions, while Figure 8(b) presents the top view of the GMD bridge with the installed SHM system and the extended part. To facilitate bridge health management, an SHM system consisting of 27 SGs and 18 accelerometers was installed to monitor the vibrational responses of this small-scale bridge.^38,42 Figure 9(a) and (b) show examples of the sensors mounted on the bridge in real world. Additionally, the GMD bridge was thoroughly examined using a terrestrial laser scanner. Phase-based method was adopted and a FARO Focus scanner was utilized. Several anchor points were established using different reference locations to aid the creation of point cloud data. Over 10 h and 12 different locations were utilized for the scanning as shown in Figure 9(c) and (d).

Figure 8.

(a) 3D depiction of the GMD bridge and the installed SHM system (units: mm). (b) 2D top view of the GMD bridge and the installed SHM system.

Figure 9.

Photos for on-site data collections: (a) and (b) present sensor network on bridge deck, (c) and (d) show the device placement for point cloud data collection.

Model construction and alignment

The development of the BIM environment for the GMD bridge began with the creation of a 3D virtual model in Revit. To recreate the reality capture, the acquired 12-point cloud data files were integrated and combined using Autodesk Recap, as shown in Figure 10(a). It is essential to eliminate extraneous aspects from the reality capture, such as vegetation, unnecessary structures, automobiles, and obstructions. The results can be further simplified depending on the amount of information and point of interest as shown in Figure 10(b) and (c), where circles indicate different laser scanning positions. The resulting point cloud model was then transferred into CloudCompare for dimension measurement. The Revit 3D model was thus created with the measured dimensions.

Figure 10.

Process of constructing BIM model from point cloud data.

To streamline the analysis, this study focused exclusively on strain data, and therefore only the SGs from the installed SHM system were modeled. Parameters relevant to the diagnostic and prognostic analyses were established to facilitate data storage and subsequent visualization. In preparation for bidirectional data mapping, the element IDs of the virtual SGs were rigorously aligned with their respective names and recorded in the MySQL database. Figure 11 illustrates the Revit model of the GMD bridge, highlighting the setup for SHM data analysis and visualization.

Figure 11.

The 3D virtual model of the GMD bridge in Revit. Sensors circled have a residual fatigue life less than 60 years.

For the interactive diagnostic analysis in MATLAB, the bridge deck was partitioned into $14, 797$ small cuboids, each with a size of $100 mm (width) \times 100 mm (length) \times 270 mm (height)$ , as shown in Figure 11. This segmentation aimed at identifying critical areas of the deck on a horizontal plane, while the strain distribution along the height of the deck was not considered in this study. Each partition was assigned an instance parameter to store diagnostic results, and the distance from each partition to the SGs was calculated and linked to its element ID.

The Dynamo scripts utilized for the distance calculation are presented in Figure 12. The coordinates of virtual sensor elements can be easily found in Dynamo with the help of standard nodes, such as Element.GetLocation. Since partitions belong to the bridge component that they are parted from and have no independent coordinates, Bounding Box nodes were adopted to locate the coordinates of the partitions. Bounding Box is the smallest possible orthogonal box that can fit the desired partitions inside. By calculating the geometrical center of such a box, the center of the corresponding partition can be identified. The distance data table, with a size of $14, 797 (row) \times 28 (column)$ , was organized with the first column containing the element IDs and the subsequent 27 columns detailing the distances to each SG.

Figure 12.

Examples of dynamo scripts for distance calculation.

Data-driven SHM process

Assuming that traffic follows the same pattern throughout the years, representative strain data collected on two independent sampling days with an interval of half a month were utilized for analysis. After denoizing, the strain data were compiled into two matrices for the subsequent diagnostic and prognostic analyses. These matrices, each with a size of $17, 280, 000 (row) \times 27 (column)$ , were referred to as $S_{1}$ and $S_{2}$ .

Diagnostic process

In order to cover most areas of the deck and avoid excessive overlapping, this study defined the diameter of the influence zone of each SG as 2 m. Thus, the parameters ( $k$ and $b$ ) in Equation (2), were determined as −0.001 and 1 respectively, to ensure that strain reaches its maximum at the SG location and decreases to 0 at a distance of 1 m from the SG. On the other hand, the standard deviation $σ$ in Equation (3) is set as $\frac{1000}{3} mm$ to ensure that 99.7% of the strain values lie within the defined circular area. With the distance table imported from the MySQL database, the SCCs $(R)$ under both the negative linear relationship ( $R_{l}$ ) and the normal distribution ( $R_{n}$ ) were calculated employing Equations (2) and (3), respectively. Given that each partition corresponds to $27$ different coefficients contributed by the $27$ SGs and 1 element ID, the matrices $R_{l}$ and $R_{n}$ each have a dimension of $14, 797 (row) \times 28 (column)$ .

To enhance computational efficiency and reduce the workload, it was crucial to minimize the sizes of the SCC matrices. Only the non-zero elements in the SCC matrices were calculated and the strain of each cuboid at each sampling point of the SGs was calculated as

s_{d (1 \times 1)} = \max (r_{(1 \times n)} (\cdot) s_{(1 \times n)})

(10)

where $r$ denotes the SCC matrix for the partition without zero elements and $n$ is the number of SGs affecting the partition. $s$ is the matrix containing strain data of one sampling point collected by the SGs affecting the partition. It is important to note that the strain of a given partition can be influenced by multiple SGs at a sampling point, which could potentially lead to an overestimation of strain values due to the cumulative calculation of the same force. To address this concern, in scenarios where overlapping influences from multiple SGs occur, only the maximum strain value calculated for that partition is considered. The 1-day maximum strain data for a given partition might result from different SGs since the traffic load could influence different gauges during the sampling time.

Prognostic process

In preparation for the application of the rain-flow counting algorithm, the strain-time data is first converted to the stress-time data (σ), followed by an essential pre-processing phase to eliminate the stress cycles that contributed insignificantly to the overall fatigue damage.⁴³ Figure 13(a) presents the processing of an 8-min strain data collected from SG 8 to locate the maximum and minimum values of each stress cycle. While these values were retained, the intermediate data points were removed as shown in Figure 13(b). The rain-flow counting algorithm was then applied to the stress data, and the resulting stress-spectrum histogram presented in Figure 13(c) was adopted for the calculation of effective cycles ( $N_{eff}$ ). Take Day 2 data collected from SG 8 as an example, following Equations (4) to (6), the effective cycles were calculated as $2.95 \times 10^{3}$ cycles and the total life in cycle ( $N_{8}$ ) was obtained as $4.86 \times 10^{7}$ cycles.

Figure 13.

Example of strain data processing: (a) Example of data processing with original data, (b) Example of data processing with stress data that contributes to fatigue cycles, and (c) Stress spectrum histogram.

To accurately calculate the residual fatigue life of the bridge in terms of years, it is essential to first determine the unknown variables in Equations (8) and (9). These variables include the traffic volume at the time when the bridge was constructed (then traffic volume) and the current traffic volume, as well as the sampling time of the SGs. For this case study, the then traffic volume $V_{p}$ was determined using a traffic estimation guideline published in 1996, which provides insights into the traffic volumes of the 1990s.⁴⁴ The then annual average daily traffic, $V_{p}$ , was assumed to be 500 in both directions with an indicative growth rate, k, of 1.5%,^45,46 which was also in accordance with the population growth rate in NSW over the last 30 years (i.e., $n = 30$ ).⁴⁷ Hence, the current traffic volume $V_{0}$ was calculated to be 780 as

V_{0} = 500 \times {(1 + 1.5 %)}^{30} ≅ 780

(11)

In this study, the strain data was collected over two distinct days, thereby establishing the sampling time for the SGs as 1 day. Consequently, the variable $t$ representing the sampling time was assigned a value of 1 in the relevant calculations.

Based on the variables identified, the total consumed cycles ( $N_{tp})$ were calculated adopting Equation (8). In terms of strain data collected on Day 2 with SG 8, $N_{p 8}$ was calculated as $2.59 \times 10^{7}$ cycles. By subtracting $N_{p 8}$ from $N_{8}$ , the residual fatigue cycles for SG 8 were identified as $2.27 \times 10^{7}$ cycles. Adopting Equation (9), the residual life of SG 8 based on Day 2 data was calculated as 18.3 years. Following the steps, $2 \times 27$ results for the residual fatigue life were obtained from the on-site collected strain data. The locations on the bridge where the residual fatigue life was calculated to be less than 60 years were presented in Table 1 and highlighted in Figure 11. In particular, the location of SG 8 emerged as the most critical part of the deck, indicating residual life estimates of 14.4 and 18.3 years, respectively, from the two sets of data. The extremely low residual fatigue life calculation for SG 8 may be due to factors such as local structural degradation or issues with sensor installation. Regardless of the cause, these calculations provide valuable information for decision-makers to review and check.

Table 1.

Residual fatigue years less than 60 years.

Locations (SG)	Residual fatigueyears (Day 1)	Residual fatigueyears (Day 2)
SG 8	14.398	18.238
SG 2	31.861	35.319
SG 4	44.143	50.596
SG 5	45.104	49.054
SG 14	45.700	50.171
SG 3	45.860	51.853
SG 9	53.546	54.242

SG: strain gauge.

Diagnostic and prognostic visualizations

With the preparations above, the visualizations of diagnostic and prognostic results were achieved along with the 3D virtual model of the GMD Bridge in Revit as shown in Figure 14(a) and (b). The strain values and residual fatigue life were mapped to their corresponding parameters of the partitions and virtual SG entities, respectively. Then, color-coded schemes were employed to represent varying levels of strain intensity and residual fatigue life duration. Additionally, as shown in Figure 14(c), real-time sensor data were visualized with the virtual 3D model as raster images, each linked to the corresponding parameters of the virtual SGs. The use of Dynamo for the real-time sensor data visualization provided detailed insights into the strain data, further enriching the understanding of the structural health of the bridge.

Figure 14.

Visualization of structural information in Revit: (a) Prognostic data visualization (Residual fatigue life), (b) Diagnostic data visualization (The most deformed area), and (c) Real-time sensor data visualization.

The visualizations of the daily maximum strain the bridge experienced are presented in Figure 15, while Figure 16 shows the distribution of the daily average strains. In both figures 15 and 16, Figure (a) and (c) depict the strain distribution derived from the strain data collected on Day 1, while Figure (b) and (d) illustrate the results from Day 2. The strain distributions in Figure (a) and (b) are based on the normal distribution rule, whereas Figure (c) and (d) are the outcomes of the negative linear relationship rule. Each figure features a zoomed-in view of the critical area, located in the left corner. Figure 15 provides the information for timely identification of anomalies, while Figure 16 presents a long-term pattern of the strains along the deck.

Figure 15.

Maximum strain distribution on the deck. (a) and (c) Strain distribution of Day 1 under normal distribution and negative linear relationship rules, respectively. (b) and (d) Strain distribution of Day 2 under normal distribution and negative linear relationship rules, respectively.

Figure 16.

Average strain distribution on the deck. (a) and (c) Strain distribution of Day 1 under normal distribution and negative linear relationship rules, respectively. (b) and (d) Strain distribution of Day 2 under normal distribution and negative linear relationship rules, respectively.

It was also observed that larger areas were highlighted in orange under the normal distribution rule, which seems particularly sensitive to large loadings and tends to yield more conservative results. In contrast, the negative linear relationship rule distributes the recorded strains over a broader range, offering a different perspective on the strain distribution. Notably, in all figures, the area near SG 8 is highlighted in orange, indicating the highest strain intensity on the deck and identifying it as the most critical area, which is consistent with the prognostic results in Table 1.

The visualization details of the prognostic results are depicted in Figure 17(a) and (b) based on the data collected on Day 1 and Day 2, respectively. In this scheme, deeper shades of green correspond to longer residual life spans. In both Figure 17(a) and (b), SG 8 is distinctly highlighted in light green, signifying the location with the shortest calculated residual life spans, and therefore is the most critical location on the deck. It is important to note that the predicted residual fatigue life does not necessarily indicate an imminent structural collapse at the end of the estimated period. Rather, it serves as a valuable indicator for determining the timing of critical maintenance and the frequency of necessary on-site inspections.

Figure 17.

Visualizations of prognostic results: (a) and (b) Showing the results obtain from Day 1 and 2, respectively.

Discussion

This study aimed at aiding bridge O&M phases monitoring through the integration of BIM techniques and data-driven SHM. The proposed methodologies were effectively demonstrated through a two-span reinforced concrete bridge as a case study. Though data analysis and 3D visualizations for diagnosis and prognosis were achieved with on-site collected strain data, limitations and challenges for the current framework still exist.

Firstly, the proposed framework adopted a relational database, which has inherent drawbacks when dealing with large datasets although it is highly compatible with both Revit environment and MATLAB. Moreover, it is also limited in processing unstructured or semi-structured data, which is common in bridge health monitoring.

Secondly, since the proposed framework was ad hoc designed for SHM data visualization and bridge local health monitoring, the diagnostic and prognostic algorithms rely heavily on field strain measurements, which may limit their applications to road bridges that are mainly subjected to cyclic traffic loading only. The generality of the effectiveness of the proposed framework for other types of bridges with different loading conditions will be the focus of future studies.

Thirdly, though the update of structural behavior such as residual fatigue life is achievable by adopting the proposed framework with new data collected by on-site sensors, challenge still exists in ensuring the dynamic updating and management of the BIM model throughout the entire lifecycle of the structure. To support the full-life monitoring of bridges and ensure the BIM model to be an accurate tool for long-term structural management, it is indispensable to dynamically integrate more critical information such as bridge damage and subsequent repairs or interventions into the BIM model over time.

Conclusion

This study introduced a framework that utilizes the BIM techniques for bridge SHM, supporting bridge diagnostic and prognostic analyses while facilitating their visualization. To achieve this, the data communication channel between the data-driven SHM process and the BIM environment was established, allowing bidirectional data mapping. Additionally, the study has proposed and customized diagnostic and prognostic algorithms specifically designed for graphical presentation within the BIM environment, thereby enabling intuitive visual assessments of bridge integrity and longevity.

The proposed algorithms were applied to a case study bridge. The BIM model was actively involved in the diagnostic and prognostic evaluations by embedding the geometrical information of the virtual model into the evaluation processes. The 3D visualization of diagnostic and prognostic data directly related the potential damage with the dimensions of the structure, improving the precision for damage localization. The visualization of diagnostic data is capable of being updated within seconds, accurately reflecting the distribution of strains across the entire deck for each sampling point of the SGs. The graphical presentation and visualization of the diagnostic and prognostic data provided timely identification of structural issues and traffic patterns and assisted in the long-term planning of maintenance schedules.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Ye Lu

Jianghua Deng

References

Chan

THT

Tam

, et al. Fiber Bragg grating sensors for structural health monitoring of Tsing Ma Bridge: background and experimental observation. Eng Struct 2006; 28(5): 648–659.

Kim

Pakzad

Culler

, et al. Health monitoring of civil infrastructures using wireless sensor networks. In: Proceedings of the 6th international conference on information processing in sensor networks, 2007, pp.254–263. Cambridge, MA: Association for Computing Machinery.

Fujino

Murata

Okano

, et al. Monitoring system of the Akashi Kaikyo Bridge and displacement measurement using GPS. In: SPIE’s 5th annual international symposium on nondestructive evaluation and health monitoring of aging infrastructure. 2000, vol. 3995. Newport Beach, CA: SPIE.

Zhao

, et al. Vibration signal-based early fault prognosis: status quo and applications. Adv Eng Inform 2022; 52: 101609.

Wang

, et al. Critical review of data-driven decision-making in bridge operation and maintenance. Struct Infrastruct Eng 2021; 18(1): 47–70.

Moallemi

Burrello

Brunelli

, et al. Model-based vs. data-driven approaches for anomaly detection in structural health monitoring: a case study. In: 2021 IEEE international instrumentation and measurement technology conference (I2MTC), Glasgow, 2021.

Guo

Shi

Gong

, et al. Statistical characteristics of modal damping of long-span suspension bridge at different wind speeds. J Vibr Contr 2024; 30(3–4): 487–501.

Vitola

Pozo

Tibaduiza

, et al. A sensor data fusion system based on K-nearest neighbor pattern classification for structural health monitoring applications. Sensors (Basel) 2017; 17(2): 417–417.

Nguyen

TQ.

A data-driven approach to structural health monitoring of bridge structures based on the discrete model and FFT-deep learning. J Vibr Eng Technol 2021; 9(8): 1959–1981.

10.

Gomez-Cabrera

Escamilla-Ambrosio

PJ.

Review of machine-learning techniques applied to structural health monitoring systems for building and bridge structures. Appl Sci 2022; 12(21): 10754.

11.

Ghasemi

Diallo

, et al. Comparative study on logical analysis of data (LAD), artificial neural networks (ANN), and proportional hazards model (PHM) for maintenance prognostics. J Qual Mainten Eng 2019; 25(1): 2–24.

12.

Wang

T-W

Yang

D-H

, et al. Data-driven modal equivalent standardization for early damage detection in bridge structural health monitoring. J Eng Mech 2023; 149(1): 06022004.

13.

Lei

Chen

, et al. Detecting and testing multiple change points in distributions of damage-sensitive feature data for data-driven structural condition assessment: a distributional time series change-point analytic approach. Mech Syst Signal Process 2023; 196: 110344.

14.

Pan

Azimi

Yan

, et al. Time-frequency-based data-driven structural diagnosis and damage detection for cable-stayed bridges. J Bridge Eng 2018; 23(6): 04018033.

15.

Wang

Liu

Z-G

, et al. Towards probabilistic data-driven damage detection in SHM using sparse Bayesian Learning Scheme. Struct Control Health Monit 2022; 29(11): e3070.

16.

Wang

Tang

, et al. Assessment and prediction of high speed railway bridge long-term deformation based on track geometry inspection big data. Mech Syst Signal Process 2021; 158: 107749.

17.

Sun

Shang

Xia

, et al. Review of bridge structural health monitoring aided by big data and artificial intelligence: from condition assessment to damage detection. J Struct Eng 2020; 146(5): 04020073.

18.

Sadhu

Peplinski

Mohammadkhorasan

, et al. A review of data management and visualization techniques for structural health monitoring using BIM and virtual or augmented reality. J Struct Eng 2023; 149(1): 03122006.

19.

Kumar

Kota

SR.

IoT enabled diagnosis and prognosis framework for structural health monitoring. J Ambient Intell Humaniz Comput 2023; 14(8): 11301–11318.

20.

Napolitano

Blyth

Glisic

Virtual environments for visualizing structural health monitoring sensor networks, data, and metadata. Sensors 2018; 18(1): 243.

21.

Catbas

Luleci

Zakaria

, et al. Extended reality (XR) for condition assessment of civil engineering structures: a literature review. Sensors (Basel Switzerland) 2022; 22(23): 9560.

22.

Al-Sabbag

Yeum

Narasimhan

Interactive defect quantification through extended reality. Adv Eng Inform 2022; 51: 101473.

23.

Shim

Yun

Song

HH.

Application of 3D bridge information modeling to design and construction of bridges. Procedia Eng 2011; 14: 95–99.

24.

Dayan

Chileshe

Hassanli

A scoping review of information-modeling development in bridge management systems. J Constr Eng Manage 2022; 148(9): 03122006.

25.

McGuire

Atadero

Clevenger

, et al. Bridge information modeling for inspection and evaluation. J Bridge Eng 2016; 21(4): 04015076.

26.

Marzouk

Hisham

Bridge information modeling in sustainable bridge management. In: ICSDC 2011: integrating sustainability practices in the construction industry, Kansas City, Missouri.

27.

Jeong

Hou

Lynch

, et al. An information modeling framework for bridge monitoring. Adv Eng Softw 2017; 114: 11–31.

28.

Boddupalli

Sadhu

Azar

, et al. Improved visualization of infrastructure monitoring data using building information modeling. Struct Infrastruct Eng 2019; 15(9): 1247–1263.

29.

Deng

Lai

, et al. Visualization and monitoring information management of bridge structure health and safety early warning based on BIM. J Asian Archit Build Eng 2022; 21(2): 427–438.

30.

Sakr

Sadhu

Visualization of structural health monitoring information using internet-of-things integrated with building information modeling. J Infrastruct Intell Resilience 2023; 2(3): 100053.

31.

Valinejadshoubi

Bagchi

Moselhi

Development of a BIM-based data management system for structural health monitoring with application to modular buildings: case study. J Comput Civil Eng 2019; 33(3): 05019003.

32.

Pertceva

Khizhnyak

Radaev

Algorithm of designing complex shape construction using automation tools (by example of Autodesk Revit, Autodesk AutoCAD and Dynamo). Transportnye Sooruženiâ 2018; 5(4): 04sats418.

33.

YY.

Hypersensitivity of strain-based indicators for structural damage identification: a review. Mech Syst Signal Process 2010; 24(3): 653–664.

34.

Wang

Modal strain energy-based structural damage identification: a review and comparative study. Struct Eng Int 2019; 29(2): 234–248.

35.

Zhou

YE.

Assessment of bridge remaining fatigue life through field strain measurement. J Bridge Eng 2006; 11(6): 737–744.

36.

Kashefi

Zandi

Zeinoddini

Fatigue life evaluation through field measurements and laboratory tests. Procedia Eng 2010; 2(1): 573–582.

37.

Holman

Liaw

PK.

Methodologies for predicting fatigue life. JOM 1997; 49(7): 46–52.

38.

Lin

STK

Alamdari

, et al. Neural network based numerical model updating and verification for a short span concrete culvert bridge by incorporating Monte Carlo simulations. Struct Eng Mech 2022; 81(3): 293–303.

39.

Glinka

Kam

JCP

. Rainflow counting algorithm for very long stress histories. Int J Fatigue 1987; 9(4): 223–228.

40.

Eljufout

Toutanji

Al-Qaralleh

Fatigue stress-life model of RC beams based on an accelerated fatigue method. Infrastructures 2019; 4(2): 16.

41.

Lopez

Data hierarchy configuration in Revit, https://www.modelical.com/en/gdocs/revit-data-hierarchy/ (2022, accessed 16 March 2023).

42.

Lin

STK

Alamdari

, et al. Field test investigations for condition monitoring of a concrete culvert bridge using vibration responses. Struct Control Health Monit 2020; 27(10): e2614.

43.

Rychlik

Simulation of load sequences from rainflow matrices: Markov Method. Int J Fatigue 1996; 18(7): 429–438.

44.

AustStab. Estimating the design traffic on stabilised pavements. 1996, p.2. North Sydney, NSW, Australia: AustStab.

45.

Kenworthy-Groen

A review of traffic growth rate calculations. In: 25th ARRB conference. 2012. Perth, Western Australia, Australia.

46.

The Bureau of Transport Economics. Assessment of the Australian Road System. Australia: Australian Government Publishing Service (AGPS Press), 1987.

47.

Angus

Trends in NSW population growth. New South Wales Parliamentary Research Services (ed.). New South Wales, Australia: New South Wales Parliament, 2019, p.17.

Building information modeling supported bridge structural health diagnosis and prognosis

Abstract

Keywords

Introduction

Methodology

Framework introduction

Diagnostic and prognostic algorithms

Diagnostic algorithm

Prognostic algorithm

Data communication and visualization

Case study

Model construction and alignment

Data-driven SHM process

Diagnostic process

Prognostic process

Diagnostic and prognostic visualizations

Discussion

Conclusion

Footnotes

Declaration of conflicting interests

Funding

ORCID iDs

References