Health Index-Based Maintenance of Prestressed Concrete Bridges Considering Building Information Modeling Application

Abstract

Bridge maintenance faces challenges regarding data management and decision-making efficiency, primarily owing to the manual processing of extensive inspection data and the absence of integrated digital solutions. This study addresses these challenges by proposing a health index (HI)-based bridge evaluation framework for prestressed concrete bridges, based on building information modeling (BIM) for inspection data integration and long short-term memory (LSTM) models for accurate deterioration prediction. The proposed framework categorizes and analyzes bridge deterioration data from inspection reports and develops a predictive LSTM model that allows quantitative bridge evaluation based on accumulated historical data. The results demonstrate that this approach enhances the efficiency and accuracy of bridge condition evaluation while enabling long-term prediction of deterioration trends. In a case study of a bridge, the bridge-level HI value decreased by 17% over 16 years, while the condition grade remained unchanged. The findings suggest that integrating BIM with HI-based bridge evaluation can support sustainable bridge maintenance strategies.

1. Introduction

Bridge maintenance generates a substantial amount of inspection data, including design materials containing bridge specifications, data on various defects observed during service, repair records, and evaluations. In conventional bridge maintenance, these data are stored in documents and manually reviewed against the maintenance standards during bridge diagnosis. Consequently, data analysis becomes time intensive, leading to low data utilization efficiency.

Building information modeling (BIM) has been introduced to address these issues by improving information management from planning to maintenance [1,2]. The primary advantage of BIM is its ability to merge an intelligent three-dimensional (3D) physical model with an environment that supports smooth interoperability across all bridge design applications. This integration enables efficient and reliable information exchange between various tools and software, promoting a unified and streamlined process for bridge design and management [3].

The applications of BIM in bridge maintenance are diverse. McGuire et al. [4] developed a BIM system for collecting and storing damage information, enabling inspectors to systematically monitor the deterioration process in real time. Tanaka et al. and Wan et al. [5,6] proposed a BIM-based bridge management system considering an Industry Foundation Classes (IFC) extension for maintenance data. The system is web-based, enabling the seamless exchange of BIM data with multiscale visualization and collaboration capabilities. Byun et al. [7] built a BIM-based bridge management system for detailed data management. They developed a web data management program that incorporated a data schema and information system linked to a modeling library and a bridge assessment process based on national maintenance guidelines. Jeon et al. [3] developed a similar BIM-based bridge maintenance system. Their study reflected the opinions of maintenance practitioners and considered the use of image data such as drone scanning. Studies focusing on managing defect data using BIM have also been conducted [8,9]. To consider defect information in a BIM environment, the International Framework Dictionaries standard was applied, and a BIM modeling library for defects was developed, which enabled the BIM-based condition rating of bridges.

Despite the potential benefits of BIM in bridge management, several challenges associated with its implementation remain, including high initial adoption costs, specialized training requirements, compatibility issues with existing systems, and the complexity of integrating BIM with conventional workflows. Existing studies incorporating BIM for bridge maintenance [5,6,10,11] have considered information exchange through IFC, primarily focusing on linking the BIM defined during the construction phase. However, from the perspective of the maintenance authority responsible for managing numerous bridges in a particular jurisdiction with limited resources, few bridges have been constructed using BIM, and establishing BIM for existing bridges incurs significant costs. Therefore, a system that enables simplified modeling based on optimized data and constructs only the minimum required level of detail (LOD) necessary for maintenance is essential.

Another challenge is the lack of a quantitative bridge assessment method that considers accumulated data. Existing bridge management systems mainly rely on the health index (HI), which is primarily derived from visual inspection data, with condition rating being the most widely adopted approach [12]. This approach has been widely adopted in developed countries including Korea, the United States, the United Kingdom, and Japan. The results from this approach have been used in maintenance decision-making [13,14,15,16,17,18,19,20]. This method is widely used owing to its convenience in reporting and determining management priorities based on bridge conditions. Previous studies [3,7] have focused on automating the condition rating process within the framework of BIM-based bridge evaluation. However, grouping data, including the time history of damage, generally results in information loss and the potential disappearance of critical features. Moreover, because the approach evaluation relies on the current state, it has limitations in predicting long-term deterioration trends and future bridge conditions. This phenomenon is similarly observed even when applying machine learning techniques. The literature [15,21,22] identified factors influencing the condition grade using long short-term memory (LSTM) to predict the condition grade of bridges based on data analyzed using multiple inspection datasets. These studies modeled the relationship between these factors and the condition grade using LSTM and reported a reasonable accuracy. However, the efficiency of such models could not be clearly verified. For example, in a study that simultaneously applied various trend prediction methods, including linear regression and LSTM, no significant differences in performance were observed [21].

To consider a more precise bridge maintenance system, Structural Health Monitoring (SHM) technologies have been actively researched [23,24,25]. These studies aim to apply IoT technologies to bridges in the context of the Fourth Industrial Revolution, enabling real-time monitoring of structural stability, which can be accomplished by processing vibration-induced data and detecting abnormal responses. Additionally, damage can be detected without visual inspection using the Eigenvalue Perturbation (EP) method. This methodology identifies the location and extent of damage by analyzing subtle changes in the dynamic properties of a bridge caused by the damage. The feasibility of this approach has been demonstrated in several studies [26,27,28]. These technologies differ from traditional HI-based evaluation methodologies in various aspects beyond simply relying on sensing data (

Table 1. Comparison between HI-based evaluation and EP-based evaluation.

Aspect	HI-Based Evaluation	EP-Based Evaluation
Purpose	Quantifies the overall condition of the bridge using a simple index	Identifies the cause and location of damage by analyzing dynamic properties
Evaluation criteria	Represents the bridge’s condition with a single index (e.g., 0–1 or percentage)	Analyzes detailed damage mechanisms based on eigenvalue and eigenvector changes
Focus	Overall performance condition and remaining service life	Localization and characterization of damage causes
Complexity	Relatively simple and comprehensive	Involves complex mathematical or physical modeling
Data processing	Uses machine learning or statistical weighting to calculate the HI	Analyzes the changes in stiffness, mass, damping, and dynamic parameters.
Accuracy	Suitable for overall condition assessment; limited in detailed damage analysis.	Enables detailed analysis of physical causes behind eigenvalue changes, offering high precision.
Application	Prioritizing maintenance and long-term monitoring of bridges.	Diagnosing damage causes, assessing urgent repair needs, and verifying models.

).

Despite the advantages of SHM and EP-based evaluation, HI-based evaluation methods continue to be used because many bridges are managed through visual inspections without sensors due to human and financial constraints. Furthermore, EP-based evaluation focuses primarily on accurately determining the current state of bridges, with limited research on predicting their future conditions [25]. From the perspective of managers overseeing numerous bridges, methods that can predict the future condition of bridges without sensors and evaluate maintenance priorities are required.

This study proposes a novel evaluation method that derives quantitative results from accumulated data, particularly for prestressed concrete (PSC) bridges. PSC bridges are among the most widely used structural types worldwide. However, numerous cases of sudden and severe damage (e.g., corrosion) and failure have been reported, highlighting the urgent need to establish effective maintenance strategies for these structures [29]. Therefore, this study focused on PSC bridges and developed a method for quickly determining maintenance priorities based on bridge maintenance records.

The remainder of this paper is organized as follows. Section 2 presents an analysis of the bridge maintenance system currently implemented in Korea, along with the method used to record inspection data. Section 3 describes the process of collecting data from inspection records and developing deterioration models using the LSTM algorithm. Accordingly, an HI-based evaluation framework is developed to facilitate the quantitative assessment of bridges, and a BIM model library is proposed to visualize the evaluation results. Section 4 presents the application of the proposed method to actual PSC bridges, including a visualization of the evaluation results in 3D models for integration with BIM, offering a solution to enhance decision-making in bridge maintenance. Section 5 presents the concluding remarks, limitations, and future work.

2. Analysis of Existing Evaluation Systems

Bridge maintenance activities in Korea are conducted according to the Korean Bridge Maintenance Guidelines [13,14]. Routine maintenance involves “regular safety inspections” conducted biannually, primarily relying on visual assessments to evaluate the functional condition and ensure that the structure satisfies the current operational standards. If significant defects are identified at this stage, the administrator implements suitable measures, such as performing more in-depth inspections. “Precise safety inspection” includes detailed visual examinations, measurements, and tests to detect any changes from the previously recorded condition and ensure that the structure satisfies the current operational requirements. Considering the current state and age of the structure, these inspections are performed at intervals of 1–4 years, with detailed records maintained on damage locations and quantities. A “precise safety diagnosis” involves meticulous visual inspections, comprehensive measurements, and tests aimed at identifying defects that may not be easily detected through regular safety inspections. This approach provides critical data for the structural conditions and safety assessments. This process is separated from other inspections by the loading test performed on the bridge to assess its safety. This method combines the findings from the condition assessments and safety evaluations to comprehensively evaluate the overall performance of bridges.

For the “precise safety inspection” and “precise safety diagnosis” discussed above, the documentation of the location and extent of damage is essential, as presented in Figure 1 and

Table 2. Defect quantity table related to Figure 1.

No.	Year	Defect Type	Quantity
			Width (mm)	Length (m)	Volume (m2)
1	2014	Efflorescence	1.20	2.30	0.36
2	2014	Efflorescence	0.60	2.00	1.20
3	2014	Efflorescence	0.60	1.00	0.60
4	2014	Efflorescence	0.50	1.50	0.75
5	2014	Efflorescence	1.20	1.20	1.44
6	2014	Crack	0.10	1.50	1.50
7	2014	Efflorescence	1.00	6.00	6.00
8	2018	Efflorescence	1.00	3.00	3.00
9	2016	Separation	0.80	0.80	0.64
10	2016	Efflorescence	0.50	3.00	1.50
11	2016	Leakage		0.40	0.40
12	2016	Efflorescence	0.80	0.80	0.64
13	2016	Efflorescence	1.00	6.00	6.00
14	2016	Efflorescence	0.40	0.40	0.16
15	2016	Efflorescence	1.00	2.00	2.00
16	2016	Efflorescence	1.00	3.00	3.00
17	2016	Efflorescence	1.00	1.20	1.20
18	2018	Efflorescence	1.00	3.00	3.00
19	2018	Efflorescence	1.00	1.50	1.50
20	2018	Efflorescence	0.60	1.00	0.60
21	2018	Leakage		1.00	1.00

. These records are maintained for each structural component, span, and defect type and are crucial for creating maintenance plans, including estimating repair volumes. Figure 2 illustrates the procedure for performing a condition rating on a bridge based on defect information. Data are aggregated by damage type to derive the condition grade and index, with the same procedure applied to each structural member and span with weighted factors, ultimately resulting in a bridge-level condition grade (from A to E) [3,7]. In Korea, the condition grade primarily dictates the maintenance prioritization for most bridges. An analysis of maintenance reports reveals that repair urgency is generally determined based solely on the current condition grade without considering the environment or damage pattern. Grade C is the most common threshold for initiating the repair. This reactive maintenance approach [29] is insufficient for forming comprehensive long-term maintenance strategies, leading to considerable performance discrepancies, even among structures that share the same condition grade.

3. Development of an HI-Based Bridge Evaluation System

The HI is a critical tool for evaluating the structural and functional conditions of bridges, including the condition rating, determined by the state of serviceability of structural components and bridges. It primarily highlights the structures that deteriorated the most within an asset inventory, prioritizing those requiring immediate repair. However, a typical limitation is that most HI systems depend on the current condition of the bridge components, limiting their effectiveness as proactive indicators of factors, such as the rate of bridge aging. Thus, a more comprehensive utilization of the inspection data generated throughout the bridge maintenance process is necessary. Initial research efforts to consider these data have focused on identifying relationships between the condition grade derived from ratings and various factors influencing bridge conditions through regression analysis to provide a basis for informed maintenance decision-making [30]. However, deterioration models developed using simple regression analysis have shown low reliability owing to numerous factors impacting bridges. To address this limitation, recent advancements have incorporated machine learning techniques. Among these, LSTM has gained widespread use owing to its ability to effectively capture the time-series characteristics of inspection data, with previous studies applying it for bridge condition grade prediction [15,21,22]. Nevertheless, these approaches remain limited by their focus on predicting condition grades alone, and results comparing regression- and machine learning-based prediction models typically have minimal discernible differences [21]. To overcome these limitations, this study proposes a new approach in which deterioration factors are used as LSTM training data to develop a deterioration model, resulting in a new HI-based bridge evaluation method, as shown in Figure 3.

3.1. Deterioration Models

A previous study introduced the LSTM algorithm to develop a carbonation prediction model for reinforced concrete slab bridges [31]. The study analyzed 242 inspection reports, resulting in 1379 carbonation data points, with 15 distinct features defined and used for LSTM training. Another study [32] analyzed 856 inspection reports related to reinforced concrete slab bridges and developed deterioration models categorized by region, bridge component, and deterioration type. This study builds on previous studies to develop an LSTM-based prediction model for various deterioration factors specific to PSC bridges, as illustrated in Figure 4. In this study, the learning model consisted of two LSTM layers, each configured with 15 input data nodes. A hyperbolic tangent function was used as the activation function for each layer, and a dropout rate of 0.5 was applied to each layer to mitigate overfitting. The predicted outcomes were subjected to a regression analysis to establish a comprehensive deterioration model. The training data comprised diverse defect data extracted from inspection reports and filtered according to the region, bridge component, and deterioration type. These data were scaled, as shown in Equation (1), and the scaled deterioration area, ${\phi }_{\left(t\right)}$, was combined with the service year, $t$, to form the dataset. Considering the inherent uncertainty in manual inspection records, outliers were excluded to reflect the overall trend. The defect inspection records used in this study exhibited high aleatory uncertainty owing to factors such as the inspector’s level of expertise, inspection methods, and differences in inspection standards, requiring several data refinement processes. as defined in [32]. The first step involved analyzing data trends, where researchers manually compared records showing abrupt changes against repair records and identified the causes of such phenomena. The second step addressed numerous unidentified anomalies during the first step by applying outlier removal algorithms, such as the interquartile range (IQR), random forest, and Bayesian neural network (BNN). Subsequently, if data for specific years were unavailable, interpolation techniques were employed to supplement the dataset for the LSTM training. Multiple data entries for the same $t$ were averaged. The trained LSTM model can predict deterioration data over a 100-year period. Power regression, as described by Equation (2), was performed to derive the final deterioration model.

$${\phi }_{\left(t\right)}^{\left(i\right)}={\displaystyle \frac{\sum _n{A}_{d\left(t\right)}^{\left(i\right)}}{A_c}},$$

(1)

$$d_{\left(t\right)}^{\left(i\right)}=\alpha t^{\beta },$$

(2)

where $t$ is the service year of a bridge when the defect data are recorded, $i$ indicates the deterioration type, $\sum _n{A}_{d\left(t\right)}^{\left(i\right)}$ is the sum of the deterioration areas with the same $t$ and $i$ within a bridge component, $A_c$ denotes the surface area of the bridge component, $d_{\left(t\right)}^{\left(i\right)}$ is $i$-th deterioration model with respect to $t$, and α and β are coefficients representing the characteristics of the degradation model.

Figure 4. LSTM learning process for the deterioration prediction models.

Figure 4. LSTM learning process for the deterioration prediction models.

3.2. Bridge Evaluation Using HI

Figure 5 illustrates the HI-based bridge evaluation process. The process begins by collecting inspection data for a particular year and aggregating the defect quantities by deterioration type, as illustrated in Figure 6. Although the procedure resembles that described in Section 3.1, the key difference is that the data are acquired only for a specific timeframe relevant to the target bridge. The aggregated quantities were then used to compute ${\phi }_{\left(t\right)}^{\left(i\right)}$ following Equation (1), and the deterioration-level HI ($DHI$) was calculated using Equation (3). The $DHI$ indicates the proximity of the current damage level to the threshold, defined by the predicted deterioration at $t=100$. In Figure 5, ${∆\phi }_{\left(t\right)}^{\left(i\right)}=0$ indicates that no new defect is detected during the most recent inspection. In such cases, the LSTM-trained deterioration model can be used to generate the $DHI$ curve, which includes regional and structural deterioration characteristics but does not directly account for the specific damage features of the target bridge. $DHI$ is essentially a ratio of the current defect to the critical defect. In other words, this value indicates the distance of a defect observed at a certain time from the threshold, and the closer $DHI$ is to 0, the more severe the defect is. In particular, $d_{\left(100\right)}^{\left(i\right)}$ is set as the threshold in this study because the design life of Korean bridge structures is set to 100 years. However, this value can be defined differently if required.

$${DHI}_{\left(t\right)}^{\left(i\right)}=1-{\displaystyle \frac{{∆\phi }_{\left(t\right)}^{\left(i\right)}+d_{(t-1)}^{\left(i\right)}}{d_{\left(100\right)}^{\left(i\right)}}},$$

(3)

where ${DHI}_{\left(t\right)}^{\left(i\right)}$ denotes the $DHI$ for the $i$-th deterioration during inspection year $t$, $d_{(t-1)}^{\left(i\right)}$ is the deterioration value derived from Equation (2) during ($t-1$), $d_{\left(100\right)}^{\left(i\right)}$ represents the threshold deterioration level of the $i$-th deterioration at $t=100$, and ${∆\phi }_{\left(t\right)}^{\left(i\right)}$ denotes the variance of the scaled deterioration area compared with the previous inspection data for each bridge component.

The subsequent step involves calculating the component-level HI ($CHI$). This procedure is similar to rating the condition at the component level based on defect data collected according to maintenance guidelines [13,14], allowing the assessment of damage levels for each component. As illustrated in Figure 1, multiple types of damage can occur within a single bridge component. In conventional condition rating systems, each type of damage is evaluated based on specific criteria, and the lowest grade is typically considered the overall rating of the component. In contrast, the HI-based bridge evaluation method provides a more detailed evaluation by considering all deterioration types to derive a quantitative index. Applying an appropriate weighting to each $DHI$ is essential because each type of $DHI$ contributes differently to the overall performance of a bridge. This weighting approach has been outlined in previous studies and listed in

Table 3. Weights of deteriorations for DHI.

Deterioration Type	Deterioration Weight	Module	Module Weight
Microcrack	0.25	Cracking defects	0.3
Macrocrack	0.375
Crazing	0.375
Separation	0.167	Materials defects	0.15
Spalling	0.167
Concrete failure	0.167
Exfoliation	0.167
Segregation	0.167
Delamination	0.167
Efflorescence	0.5	Other types of defects	0.15
Leakage	0.5
Exposed rebar	1	Corrosion defect	0.4

[32]. Based on these weights, the $CHI$ can be calculated using Equation (4).

$${CHI}_{\left(t\right)}^{\left(j\right)}=\begin{array}{c}0.3\left(0.25·{DHI}_{\left(t\right)}^{\left(Micro crack\right)}+0.375·{DHI}_{\left(t\right)}^{\left(Macro crack\right)}+0.375{DHI}_{\left(t\right)}^{\left(Crazing\right)}\right) \\ +0.15\left(\begin{array}{c}0.167·{DHI}_{\left(t\right)}^{\left(Separation\right)}+0.167·{DHI}_{\left(t\right)}^{\left(Spalling\right)}\\ +0.167·{DHI}_{\left(t\right)}^{\left(Concrete failure\right)}+0.167·{DHI}_{\left(t\right)}^{\left(Exfoliation\right)}\\ +0.167·{DHI}_{\left(t\right)}^{\left(Segregation\right)}+0.167·{DHI}_{\left(t\right)}^{\left(Delamination\right)}\end{array}\right)\\ +0.15(0.5·{DHI}_{\left(t\right)}^{\left(Efflorescence\right)}+0.5·{DHI}_{\left(t\right)}^{\left(Leakage\right)})\\ +0.4·{DHI}_{\left(t\right)}^{(Exposed rebar)}\end{array}$$

(4)

where $j$ denotes the bridge components (slab, girder, pier, or abutment).

The final step involves deriving the bridge-level HI ($BHI$) by applying weights to the CHI, as shown in

Table 4. RC slab bridge component weights for CHI.

Bridge Component	Weight
Slab	0.43
Girder	0.29
Pier	0.18
Abutment	0.18

. This is performed following a method similar to that used for calculating $CHI$ from $DHI$ with assigned weights. The maintenance guidelines in South Korea [13,14] provide weights for each bridge component for the condition rating. Component-specific weights for PSC bridges were adopted in this study. Based on the weights (${\omega }_j$), $BHI$ can be calculated using Equation (5).

$${BHI}_{\left(t\right)}=0.64·{CHI}_{\left(t\right)}^{\left(Slab\right)}+0.18·{CHI}_{\left(t\right)}^{\left(Pier\right)}+0.18·{CHI}_{\left(t\right)}^{\left(Abutment\right)}$$

(5)

3.3. 3D Modeling

The 3D model discussed in this study was introduced to suggest a method for incorporating the $BHI$ concept into BIM-based bridge maintenance. As shown in Figure 7, a modeling library was created using the minimum required data. Modeling was performed using Autodesk Revit 2023 [33], a commercial BIM design software, with the library for the five primary components of PSC bridges implemented through the family function of Revit. The essential variables required for bridge modeling include bridge length, number of spans, width, and height. These variables, combined with standard Korean design drawings, enable automated generation of detailed geometric models. This approach enables managers to create bridge models at an LOD of 200 with minimal data input, even if the resulting model differs from the actual bridge details owing to the limited geometric information. However, for bridge maintenance, developing a system for accumulating and using maintenance data by component is more critical than achieving precise visual accuracy. Thus, this method is both practical and cost-effective.

4. Case Application

Inspection reports for 134 PSC bridges in Seoul were collected to develop the deterioration model. Each bridge had an average of five inspection reports, and the data were sorted by the bridge component and deterioration type according to the inspection year, as illustrated in Figure 6. A sequential approach involving the interquartile range, isolation forest, and BNN was used to filter outliers, with the BNN trained using deterioration data within the 1–σ range. The final dataset from the inspection reports spanned a maximum input of 51 years and was used to train the LSTM model, as shown in Figure 4. The TensorFlow library of Python was used for the LSTM model learning, and two LSTM layers were used with 15 input data nodes. The activation function was a hyperbolic tangent; a dropout layer of 0.5 was added to each layer to prevent overfitting. To construct the deterioration model, 100 years of data were required. Data for the remaining 49 years were predicted using the trained LSTM model. The actual (years 1–51) and predicted (years 52–100) datasets were subjected to power regression, as outlined in Equation (2), resulting in 48 distinct deterioration models (4 bridge components × 12 deterioration types). Figure 8 shows examples of six types of deterioration models (microcracks, macrocracks, crazing, efflorescence, leakage, and exposed rebar) for the slabs and piers. An analysis of these models reveals that bridges in Seoul tend to experience rapid microcrack development in both components. In contrast, exposed rebar damage is comparatively minimal. For slabs, damage types, such as efflorescence, macrocracks, and crazing, are the most prevalent after microcracks. For piers, common types of damage include crazing, leakage, and efflorescence.

The evaluation process was applied to five bridges in Seoul, South Korea, with basic information for each bridge provided in

Table 5. Basic information of case study bridges.

Name	Construction Year	No. of Inspection Reports	Width (m)	Length (m)	Height (m)	No. of Spans	Maximum Span Length (m)
Bridge A	1999	7	18	510	9	15	18
Bridge B	1979	6	30	120	12	4	30
Bridge C	1995	5	50	175	3	7	25
Bridge D	1975	5	34	90	6	5	25
Bridge E	1971	6	40	60	3	3	20

. The number of defects for each of the five bridges was calculated by analyzing 29 inspection reports, with data points marked by the inspection year as the time stamp; Figure 9 illustrates the accumulated defect quantity data. Considering the large variation in the damage quantities across different damage types, a logarithmic scale was used to represent the data. The category “others” represents the combined quantity of six deterioration types associated with “material defects”, as detailed in Table 3. An analysis of the defect distribution reveals that overall, the number of crack-related defects was higher than those of other types, except for the abutment, where leakage-related defects exceeded crack-related defects. These plots show the cumulative defect quantities, indicating that the bridge size significantly affects the results. Consequently, a high defect quantity does not necessarily indicate poor bridge conditions, and a precise evaluation requires deriving the HI.

The collected defect quantities were scaled according to Equation (1), and the $DHI$, $CHI$, and $BHI$ were calculated following the procedure outlined in Figure 5. In addition, deterioration models for PSC bridges were applied to derive the HI curve over the time range $\mathrm{t}=0–100$ with ${∆\phi }_{\left(t\right)}^{\left(i\right)}=0$. Figure 10 shows the derivation results of the $CHI$ values and $CHI$ curves for each bridge component, obtained using Equation (4). The $CHI$ curve represents the deterioration characteristics of bridges in Seoul, illustrating the degradation levels in relation to the service years. Accordingly, for bridges such as Bridge A, in which the $CHI$ value closely follows the slope of the curve, the deterioration rate appears to be consistent with that of other regional bridges, suggesting that immediate maintenance may not be required. This finding contrasts with that shown in Figure 9, where Bridge A exhibits a higher damage quantity than the other bridges. By contrast, for Bridges C and D, Figure 10c,d show several $CHI$ values falling below the $CHI$ curve, indicating a higher-than-average deterioration relative to the other bridges in the area. This suggests that these bridges warrant a higher maintenance priority.

Figure 11 illustrates the $BHI$ values and the curve derived using Equation (5) to calculate the $CHI$ values and curves. To assess the differences between the results of this study and the conventional condition ratings, the condition grades documented in the inspection reports were reviewed, as listed in

Table 6. Condition rating results of the case study bridges.

Report Number	Condition Grade
	Bridge A	Bridge B	Bridge C	Bridge D	Bridge E
#1	B	B	A	B	B
#2	B	B	B	B	B
#3	B	B	B	B	B
#4	B	B	B	B	B
#5	B	B	B	A	B
#6	B	B	-	-	B
#7	B	-	-	-	-

. Although most bridges indicate ongoing deterioration, most of them were assigned a condition rating of “B”, suggesting that no special maintenance measures were required. Bridge D exhibits the $BHI$ values significantly deviating from the $BHI$ curve and decreased by 17% across 16 years, indicating a potentially higher maintenance priority compared with the other bridges. Notably, Bridge D received an “A” grade during the final inspection year, as repairs had removed visible signs of damage. However, the repairs primarily involved superficial crack sealing, which did not address the underlying deterioration causes. In such cases, the defect quantities on the bridge may recur similarly to previous records, likely resulting in a condition rating of “B”. Nevertheless, the $BHI$ assessment indicates that Bridge D exhibits higher deterioration progression than the other bridges. Consequently, the $BHI$ assessment procedure offers a distinct perspective on bridge maintenance, providing a new basis for prioritizing maintenance efforts.

A Revit add-in module, as shown in Figure 12, was developed to integrate the previously collected defect quantity data and outputs from the HI-based bridge evaluation process into BIM. When bridge evaluation results are visualized, the $CHI$ values, expressed as decimal points, increase the complexity for bridge managers to intuitively understand their significance. Therefore, the $CHI$ values were transformed by substituting into $d_{\left(t\right)}^{\left(i\right)}$ in Equation (2) to reverse-calculate $t$, thereby determining the ages of the deteriorated components. Parameters α and β were derived from the $CHI$ curve. These calculated ages enabled bridge managers to intuitively identify which components or bridges were older, and this information was integrated into the visualization through a dedicated algorithm within the HI-based evaluation system. In addition, to predict changes in components over time, the algorithm incorporated projections of $CHI$ values 10, 20, and 30 years into the future based on the slope of the $CHI$ curve. This calculation was part of the HI-based bridge evaluation process, and the results were recorded in an .xlsx-based template. In this system, the 3D model used for visualization must first be generated using the model library shown in Figure 7. The .xlsx file was then imported into the add-in module. After importing, the add-in module is displayed in blue if the age from the $CHI$ curve is higher than that from the $CHI$ value; otherwise, it is displayed in red, with the intensity increasing as the difference increases. Detailed data on the defect quantities can be obtained by accessing the attribute information stored in each component. In addition, the module included a feature that showed predictive results for 10, 20, and 30 years in the future, facilitating a quick assessment of components showing rapid deterioration. This study could provide quantitative support for bridge management strategy formulation and contribute to rational and proactive maintenance practices.

5. Conclusions

This study presents an HI-based maintenance framework for PSC bridges that integrates BIM- and LSTM-based deterioration modeling to address inefficiencies in conventional bridge maintenance processes. The proposed methodology used inspection data from the inspection reports of 134 PSC bridges in Seoul as a case study, and 48 distinct deterioration models were derived using LSTM learning and power regression. For each of the five case study bridges, the collected defect data and the derived deterioration models were converted into $DHI$ values and $DHI$ curves. Moreover, $CHI$ values were calculated by incorporating the assigned weights. This approach enabled the quantitative identification of bridge components exhibiting higher levels of defect occurrence and analysis of defect progression patterns. Furthermore, $BHI$ values were calculated by factoring in weights across the components. Subsequently, a comparative analysis with conventional condition rating records was performed. The findings revealed that while condition grades remained consistent over the service years, the $BHI$ provided a valuable basis for prioritizing maintenance efforts. A Revit add-in module was developed to visualize the outcomes of the HI-based evaluation process, providing bridge managers with an intuitive tool for assessing bridge conditions and accessing essential information. The findings indicate that this approach significantly improves the accuracy and efficiency of bridge evaluation, addressing the limitations of conventional methods that primarily rely on manual processing and static data. Visualizing HI assessments within BIM-based 3D models offers a more comprehensive view of bridge conditions and promotes proactive maintenance strategies.

This study was conducted on PSC bridges designed based on design trucks specified in the road bridge design standards, which typically have span lengths of 20–50 m. The findings of this study apply to such bridges; however, for special cases or bridges designed with different types of loads (e.g., train loads), it would be more reasonable to separately collect inspection records for these bridges and derive dedicated deterioration models. Although this study focused specifically on PSC bridges, the proposed methodology can be extended to other bridge types provided that adequate data are available. However, owing to the heavy reliance of this method on data, data reliability is essential, and preprocessing steps are required to minimize human error. The inspection data used in this study showed significant variability, and preprocessing was time intensive. Future work should focus on refining the deterioration models to include additional environmental and material variables, expanding the validation efforts with larger datasets, and exploring cost-effective solutions to facilitate broader adoption across different bridge types and geographical regions.