Impact of Concrete Sealer and Salt Usage on Concrete Bridge Deck Condition and Life Cycle Cost

Abstract

The objective of this study is to analyze bridge deck condition deterioration and evaluate the impact of concrete sealer and salt usage on deck condition and life-cycle cost. To achieve this goal, machine learning models were built to predict the evolution of bridge deck rating. The deck maintenance history shows that the average bridge age at deck overlay and deck replacement is around 25 and 50 years, respectively. Deck overlay can improve deck condition from an average rating of 6.3 to 7.1, and deck replacement can efficiently recover deck condition from an average rating of 5.3 to 8.5. The effect of concrete sealer on bridge deck condition is only observable at the stage before the first overlay, indicating that concrete sealer may not be effective over the long term. More usages of prewet salt and salt brine in anti-icing result in slightly higher deck condition ratings, while more dry salt in deicing presents slightly lower deck condition ratings, indicating the benefits of salt brine over dry salt. When concrete sealer is applied every 2 or 4 years, it can help extend the service life of the bridge deck by around 1~2 years. If concrete sealer is applied every 12 years, a 6% reduction in life cycle cost could be achieved.

1. Introduction

Cold-weather regions face challenges in preserving bridges that are subject to harsh winter conditions and the application of chemical deicers [1]. With over 600,000 bridges spanning state and local roadways in the U.S., bridge infrastructure is crucial for ensuring the safety and efficient movement of people and cargo. However, balancing increased operational demand with a constrained budget highlights the critical need to extend the service life of these structures, while minimizing long-term maintenance costs [2]. This need is especially urgent given the widespread use of deicing chemicals in winter maintenance.

It is well-documented that deicing and anti-icing salts contribute to the deterioration of bridge infrastructure, affecting both the concrete paste and the reinforcing steel [3,4]. Chloride-based chemicals such as sodium chloride (NaCl), calcium chloride (CaCl₂), and magnesium chloride (MgCl₂) are commonly used for snow and ice control, but they are also primary inducements in promoting corrosion of embedded steel reinforcement [5]. Corrosion of reinforcing steel reduces the load-carrying capacity of bridge structures and compromises their safety. In addition, specific chemical reactions between deicing salts and the phases in cement paste, such as the formation of deleterious compounds like magnesium silicate hydrate (M-S-H), further accelerate concrete damage [6].

To protect concrete bridges from salt-induced damage, highway agencies have adopted various maintenance and preservation practices, such as regular cleaning/wash of bridge deck and joints, sealing concrete deck and parapet, replacing joint seal, and applying deck overlays (thin polymer epoxy, asphalt with waterproof membrane, concrete overlays) [7]. It has been reported that bridges designed for a 100-year life expectancy should have decks that last 55 years with progressive preservation activities. Protective paint coatings should have a service life of 25–40 years for the protection of structural steel. Sealing decks every 3 to 5 years at a minor cost can delay deck deterioration by 10–12 years [8].

Several previous studies have highlighted the efficacy of bridge deck preservation strategies. Ni et al. indicated that crack sealers are essential for extending the service life of concrete bridge decks. High-molecular-weight methacrylate (HMWM) and methyl methacrylate (MMA) sealers offer superior performance in crack penetration and sealing, while epoxy sealers excel in bond strength and freeze–thaw resistance [9]. Xia et al. developed a new self-stratifying waterproof adhesive layer coating for steel bridge decks to address durability and cost shortcomings in existing systems. The coatings exhibited excellent heat resistance, low-temperature flexibility, mechanical properties, water impermeability, and UV aging resistance [10].

Tabatabai et al. demonstrated that the proper application of a thin polymer overlay on well-maintained bridge decks (no significant chloride contamination, corrosion, and/or deck surface defects) could achieve a life expectancy of 7 to 15 years [11]. Bektaş et al. concluded that the most cost-effective preservation strategy overall is to seal decks at a General Condition Rate (GCR) of 7–8, applying two thin-polymer overlays when GCR drops to 7, and applying concrete overlay when GCR drops to 6, followed by resealing when GCR drops to 7 [12]. These preservation strategies have shown significant cost savings while maintaining structural integrity. Cramer et al. highlight the effectiveness of thermal-sprayed titanium and zinc anodes in galvanic cathodic protection systems for mitigating corrosion in coastal concrete bridges. The use of stainless-steel reinforcement combined with zinc-hydrogel or catalyzed thermal-sprayed titanium anodes demonstrates a service life exceeding 25 years without significant anode polarization [13].

Bridge asset management programs play a pivotal role in ensuring the longevity and safety of bridge infrastructure [14]. Regular review and analysis of bridge performance data allow agencies to not only implement preservation and rehabilitation strategies but also assess their effectiveness, enabling adjustment/optimization based on bridge conditions and operational requirements [15].

However, few studies have explored the integration of bridge performance data with winter maintenance records to evaluate the combined effects of deicing salts and concrete sealers on bridge deck performance and life-cycle cost [16,17,18]. Previous research has primarily focused on individual preservation techniques or material performance (e.g., sealers, overlays, or cathodic protection) [19] without addressing the broader interplay between salt application quantities, sealer effectiveness, and deck condition ratings. Additionally, limited work has provided actionable insights for optimizing preservation efficacy and cost-efficiency through life-cycle cost analysis (LCCA), leaving a critical need for more holistic approaches in this area.

Building on the current research gaps, this study leverages machine-learning technology to analyze concrete bridge deck performance data, focusing on predicting deck condition ratings based on the effects of concrete sealers and deicing salts. The methodology integrates bridge management and winter maintenance data to explore correlations between salt application methods (e.g., solid salt, salt brine, and quantities) and deck performance. Additionally, it evaluates the role of concrete sealers in enhancing deck condition ratings and conducts LCCA to assess the benefits of varying sealer application frequencies.

2. Data Collection

Bridge information and condition data were provided by one state Department of Transportation (DOT) in the U.S. The database contained bridge location, traffic data, bridge configurations, deck geometry, construction and maintenance history and costs, inspection history, and National Bridge Inventory (NBI) component condition ratings.

Bridges across the state are categorized based on main span materials such as concrete, steel, timber, and others. The wearing surface of the bridge deck may not match the main span material or deck surface as recorded, since it varies by different types of overlays and surface treatments. To investigate the deck performance of concrete based bridges, those with main span material of concrete were extracted and further filtered based on typical types of wearing surface, including concrete, integral concrete, low slump concrete, epoxy overlay concrete, polyester polymer concrete, micro silica modified concrete, latex concrete, and the ones without surface treatment. The filtered bridge data were organized and processed to eliminate the records showing unreasonable values. After filtering, the bridge database had a total of 8412 bridges with available bridge information and annual records of deck maintenance and preservation activities and deck condition ratings. Statistical analysis of the selected data shows that most of the bridges were constructed within the past 70 years. The average bridge ages at deck overlay and replacement are around 25 and 50 years, respectively.

The bridge deck condition is evaluated by the NBI condition rating method with a range from 0 (failed condition) to 9 (excellent). Based on the bridge database, the annual deck condition ratings were combined with construction and maintenance historical data. The effect of maintenance activities on annual deck conditions was analyzed. Figure 1 shows the bridge deck ratings before and after maintenance activities. The influences of deck replacement and overlay on deck rating were both significant. Deck replacement can efficiently recover deck condition from an average rating of 5.3 to 8.5; while overlay can improve deck condition from an average rating of 6.3 to 7.1. For crack sealing, deck repair, and patching, no apparent improvement in deck condition ratings was noticed.

Annual salt usage data were obtained from the winter maintenance program provided by the same state DOT. The salt usages include three types of salt: prewet salt, dry salt, and salt brine. Figure 2 shows the distributions of annual average salt usages (pounds per year) at different route segments in the state. The results show that salt usage has large variations across different locations.

3. Machine Learning Model for Predicting Deck Condition

3.1. Random Forest Algorithm

Machine learning allows data scientists to develop computerized programs to accurately predict outcomes without using traditional manual statistical methods. The prediction of machine learning is realized by relying on typical algorithms to generate new output values based on input datasets.

Based on evaluation and comparison among commonly used algorithms, including Random Forest (RF), Gradient Boosting (GB), Artificial Neural Networks (ANN), and Support Vector Regression (SVR), the RF algorithm is used with the highest accuracy in this study due to its ensemble-based structure and inherent capacity. Random Forest is a popular machine learning algorithm that produces the output by combining the results from multiple randomly selected decision trees. As shown in Figure 3, the development of the random forest machine learning model is started by splitting the main database into a training dataset (25%) and a testing dataset (75%). The training dataset is used for generating decision trees to build the random forest model with numerous decision trees. All decision trees are generated based on Bootstrap theory, by which the datasets and variables used in each decision tree are randomly selected from the training dataset. Repeated selection is allowed. Each node in decision trees is split based on Gini impurity, which is a factor representing the impurity of the dataset at each node. The variable with the lowest Gini impurity is used as the top root node. The Gini impurity is calculated as in Equation (1)

$$Gini impurity=1-{\sum }_{i=1}^n{p_i}^2$$

(1)

where $p_i$ is the percentage of data with the same response at that node.

After the model training and establishment, the variable importance can be traced to present the relative influence of each variable in the training dataset. It is calculated based on Equation (2)

$$I_i={\displaystyle \frac{1}{F}}{\displaystyle \sum _{j=1}^F}{\sum }_{k=1}^N{p}_k{∆}_{gini}$$

(2)

where $I_i$ is the importance of the variable $x_i$; $F$ is the total number of decision trees in the random forest model; $N$ is the number of nodes using the variable $x_i$; $p_k$ is the percentage of data with the same response at that node; ${∆}_{gini}$ is the change of Gini impurity between that node and the nodes in the sublayer.

The testing dataset is used to verify the model’s efficiency, presenting the prediction accuracy. The model prediction accuracy is calculated as in Equation (3)

$$Accuracy={\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\displaystyle \frac{\left|y_i-{\widehat{y}}_i\right|}{y_i}}$$

(3)

where N is the number of testing datasets; $y_i$ and ${\widehat{y}}_i$ are actual and predicted responses, respectively.

3.2. Model Development and Verification

The Random Forest algorithm was used to predict bridge deck ratings. The input parameters included bridge/deck age, application time of different preservation and maintenance activities, average daily traffic (ADT), annual average salt usage, and general bridge attributes (such as length, deck lane, deck materials).

Considering that major maintenance activities such as overlay and deck replacement have dominant influences on deck condition, the entire service life of the bridge deck was classified into four stages, as shown in Figure 4. The four stages include the following:

(1)

Stage 1 is the period from initial construction to the first overlay activity.

(2)

Stage 2 is the period from deck overlay to the first deck replacement.

(3)

Stage 3 is the period from the first deck replacement to the next overlay.

(4)

Stage 4 covers the rest of the period until the end of service life, which may contain multiple overlay activities.

Figure 4. Bridge deck service life prediction at different stages (green arrows indicating minor maintenance and red arrows indicating deck overlay or replacement).

Figure 4. Bridge deck service life prediction at different stages (green arrows indicating minor maintenance and red arrows indicating deck overlay or replacement).

Based on the classified life stages, four Random Forest models were developed separately at each stage. The input variables of bridge and deck age were time factors that respectively indicate the age of the bridge and deck since initial construction or deck replacement. The overlay age indicated the years since the last overlay was applied. The application of concrete sealer (silane/siloxane) was input as a binary index (0/1). The average daily traffic values were input as ADT of cars and trucks, respectively. The salt usages were input based on different salt types (prewet salt, dry salt, and salt brine). Other inputs of bridge deck attributes included deck length, main span length, overburden depth, main span materials, deck lane number, and so on.

The general training results of RF models are listed in

Table 1. Testing results of RF machine learning models.

Stages	MAE	R-Square	RMSE	Accuracy	Data Number
1	0.40	0.72	0.30	94.25	74,410
2	0.38	0.74	0.33	93.69	13,012
3	0.27	0.81	0.16	96.04	4717
4	0.18	0.93	0.06	97.15	152

. The performance factors include mean absolute error (MAE), R-square, root mean square error (RMSE), and accuracy as shown in Equation (3). According to the training results, all four RF machine learning models show good accuracy for predicting bridge deck ratings with R-square values of 0.72–0.93.

Figure 5 shows the importance ranking of input variables from RF models. Only variables with importance values greater than 0.05 are presented. It shows that bridge age and deck age show relatively higher importance compared to other input variables. Overlay age shows significant importance only at stages 2 and 4. The average daily traffic of cars (ADT car) presents the third importance rank in stages 1, 3, and 4, and shows the secondary importance in stage 2. The average daily traffic of trucks (ADT truck) shows relatively low importance at all stages. The importance of the three types of salt usage is similar. As for other bridge deck features, deck length and main span length show higher importance among these deck features, except in stage 4, in which the overburden depth has higher importance. It is noted that the importance of concrete sealer is low (<0.05), but it is included in the input variables for further analysis.

Correlation analysis was not conducted in this study due to the following considerations. Although certain variables, such as bridge deck age versus bridge age versus overlay age, average daily traffic (ADT) of cars versus trucks, and bridge length versus main-span length, may exhibit certain correlations, such correlations do not negatively affect the accuracy of machine learning predictions. RF algorithms do not inherently depend on assumptions of linearity or non-linearity, but instead effectively capture comprehensive patterns and interactions within data, benefiting from correlated inputs. Additionally, this study specifically analyzes the impacts of three distinct deicing salts and one concrete sealer, each independently applied to separate bridges at different locations and timeframes, thus ensuring inherent independence among these treatment-related input variables.

Figure 6 presents two examples of predicted deck ratings for the 100-year service life of a bridge using the inputs of two recorded bridges in the database. The prediction model shows good capability to predict the variations of deck condition that match well with the actual data. The effects of overlay and replacement on deck condition ratings can be clearly observed.

3.3. Prediction of Bridge Deck Condition

To investigate the effects of certain input variables on deck condition ratings, further analysis was conducted based on the developed RF machine learning models.

Table 2. Prediction scenarios of bridge deck conditions.

Prediction Scenario	Factor	Expected Output	Prediction Assumptions
1	Concrete sealer effect	No concrete sealer vs. concrete sealer	Predicted service life: 100 years Input data: bridges with random attributes based on the database. Preservation and maintenance schedule was analyzed based on the criteria below. Overlay is applied when the deck rate is below 6.3. Replacement is triggered when the deck rate is below 5.2. One overlay is applied before replacement and multiple overlays may be applied after replacement.
2	Salt effect (Prewet salt, dry salt, and salt brine)	Less vs. moderate vs. more salt (25th percentile vs. average vs. 75th percentile)

presents the design of two prediction scenarios. For each scenario, bridge attributes such as lane number, length, deck materials, etc., were randomly assigned based on the database and input into the model for prediction. The effects of concrete sealer and salt usage on deck condition ratings were analyzed as follows: (1) for concrete sealer, the comparison was between bridges with and without sealer applications; (2) for salt usage, the salt effect was represented by comparing predicted deck ratings of bridges with less, moderate, and more salt usages (including three types of salts).

Based on the analysis of bridge deck ratings in the database, the maintenance schedule was determined based on the rule that deck overlay was applied when the rating was below 6.3 and deck replacement was triggered when the rating was below 5.2. It was noted that one overlay was applied before the first deck replacement, while multiple overlays might be applied after replacement during the 100-year period.

Figure 7 shows the prediction results of three bridges in each case based on the inputs in Table 2. For the first bridge in Figure 7, the effect of concrete sealer on bridge deck condition is observable at stage 1. Although the predicted deck condition ratings with and without concrete sealer show different trends at the beginning of stage 2, the deck replacement is triggered in almost the same year. After deck replacement, which recovered the condition rating to around 8.0, the bridge with concrete sealer shows a slightly higher rating. For the other two bridges in Figure 7, the predicted deck condition ratings with concrete sealer are slightly higher than those without concrete sealer at stages 3 and 4. The results indicate that applying concrete sealer may better preserve deck ratings, but the effect is not significant.

The effects of different types of salt usage on predicted deck ratings are shown in Figure 8. According to the predicted deck condition ratings, only minor differences could be observed for bridge decks with different salt usages. More usages of prewet salt and salt brine in anti-icing result in slightly higher deck condition ratings, while less dry salt usage in deicing presents slightly higher deck condition ratings. This finding is consistent with the laboratory experimental results on the effect of deicers on concrete durability reported in a previous study [20].

4. Life Cycle Cost Analysis

4.1. Orthogonal Analysis of Factors Affecting Deck Rating

Previous analyses found that the influence of concrete sealer and salt usage on bridge deck condition ratings is not significant and varies a lot among different bridges. Thus, orthogonal analysis was conducted to further quantify the influence of concrete sealer and salt usage. Orthogonal analysis is a mathematical method that is usually used for investigating the effects of multiple factors with levels in a simplified way [21]. For instance, to fully conduct experiments with four factors and two levels for each factor, it requires 2⁴ = 16 tests. However, the tests can be reduced to eight tests with an orthogonal design table.

Table 3. Orthogonal table of prediction model inputs.

Factor	Sealer	Prewet Salt (lb)	Dry Salt (lb)	Salt brine (lb)
1–1	1 (Yes)	1 (562)	1 (43, 146)	1 (99)
1–2	1	2 (44, 150)	2 (224, 420)	2 (1706)
1–3	1	1	2	2
1–4	1	2	1	1
1–5	2 (No)	2	1	2
1–6	2	1	2	1
1–7	2	2	2	1
1–8	2	1	1	2

shows an orthogonal design table listing prediction inputs for four factors and two levels. The four factors are concrete sealer, amount of prewet salt, amount of dry salt, and amount of salt brine. The bridge deck with and without concrete sealer treatments is denoted by levels 1 and 2, respectively. The salt usages at the 25th and 75th percentile of the values as recorded in the database are set, respectively, as levels 1 and 2. For each case, the prediction was conducted for 100 years, and multiple overlays were allowed at stage 4 to maintain a deck rating above 6.3. The maintenance schedule is determined based on the rule that deck overlay is applied when the rating is below 6.3 and deck replacement is triggered when the rating is below 5.2.

The prediction results of eight cases are plotted in Figure 9 to compare the service life at stages 1–3 and the number of overlays at stage 4. The concrete sealer is applied every 2 or 4 years, and the results are similar. The results are averaged to show the effect of the level at each factor. As for concrete sealer, the bridge decks with concrete sealer treatment show longer service life at stages 1–3 and require fewer overlays at stage 4. The analysis results indicate that applying concrete sealers can help extend the service life of the bridge deck by around 1~2 years at stages 1–3 and reduce the number of overlays at stage 4.

As for salt usage, dry salt presents insignificant influence on the service life of stage 1, while less dry salt can help extend the service life of stages 2 and 3 by around half a year and reduce one overlay at stage 4. As for the other two types of salts, more usage of these salts shows a positive effect on extending service life by around 1~2 years and reducing one overlay at stage 4. Since the effect of salt usage on bridge deck condition ratings is not significant and the cost data of deicing salts on each bridge are unavailable, only the effect of concrete sealer is considered in the life cycle cost analysis.

4.2. LCCA Parameters and Results

Life cycle cost analysis (LCCA) is a systematic tool used to assess the total cost of owning a facility or running a project with flexibility and comprehensiveness. LCCA can be used to calculate all significant and relevant costs over the total life cycle of the bridge. LCCA is usefully applied to determine a bridge design alternative that will fulfill the project objective at the lowest overall cost while retaining a satisfactory service level and performance.

The total cost of the bridge deck during its service life mainly consists of initial construction cost, routine inspection costs, maintenance costs, demolition cost, and residual value [22], which can be expressed as Equation (4)

$${LCC}_{NPV}=C_{ic}+\sum _{i=1}^{n_{ri}}{\displaystyle \frac{C_{ri}{t}_i}{{(1+r)}^{t_i}}}+\sum _{k=1}^{n_{mt}}{\displaystyle \frac{C_{mt}{t}_k}{{(1+r)}^{t_k}}}+{\displaystyle \frac{C_d}{{(1+r)}^T}}-{\displaystyle \frac{R_v}{{(1+r)}^T}}$$

(4)

where ${LCC}_{NPV}$ is the total cost represented by Net Present Value (NPV); $r$ is the monetary discount rate; $C_{ic}$, $C_{ri}$, $C_{mt}$, $C_d$ and $R_v$ are costs of different activities: initial construction, routine inspection, maintenance, demolition, and residual value, respectively; $n_{ri}$ and $n_{mt}$ are number of corresponding activities during the investigated period; $T$ is the investigated service life.

Based on the NPV, the equivalent annual cost can be calculated as an indicator for comparison of the costs of bridge decks with different life periods. The equivalent annual cost is calculated as shown in Equation (5)

$$EUAC={\mathrm{L}\mathrm{C}\mathrm{C}}_{\mathrm{N}\mathrm{P}\mathrm{V}}{\displaystyle \frac{\mathrm{r}}{{1-(1+\mathrm{r})}^{-\mathrm{T}}}}$$

(5)

where $EUAC$ is the equivalent uniform annual cost, which equals ${LCCA}_{NPV}/T$ when the monetary discount rate is zero.

Table 4. Cost data of maintenance activities for bridge decks.

Cost Data Items		Cost, $/m2	Cost, $/SF
Major maintenance	New deck/Deck replacement	1130	105.0
	Deck overlay	495	46.0
Minor maintenance	Deck repair	20	1.9
	Crack sealing	15	1.4
Concrete sealer		8	0.7
Routine inspection		2	0.2

summarizes the cost data of different maintenance activities of concrete bridge decks used in LCCA. The costs of a new deck, deck replacement, and overlay are extracted from the existing literature [23]. The new deck cost and deck replacement cost are regarded as the same, $105/SF for a concrete bridge deck. The deck overlay cost is $46/SF.

Typical minor maintenance methods with corresponding costs are extracted and calculated from the bridge database, which are $1.9/SF for deck repair and $1.4/SF for crack sealing. The unit cost of concrete sealer is obtained from the communication with the state DOT, which is $0.7/SF. All minor maintenance and concrete sealer costs are unit costs based on the total area of deck surface, in keeping with the same unit as deck replacement and overlay in LCCA. The cost of routine inspection is $0.2/SF based on the data in the literature [24].

Table 5. Application intervals of deck preservation.

Preservation Activities	Application Intervals (Years)
Routine inspection	2
Deck clean, sweep, and drains	2
Deck crack sealing	2~6
Deck repair and patching	2~6

lists the application intervals of deck preservation based on the literature and observations in the bridge database. The routine inspection and deck clean, sweep, and drains are regularly applied every 2 years. Typical minor maintenance, such as crack sealing and deck repair and patching, is applied every 2–6 years. The application interval of minor maintenance is set as 4 years in LCCA.

Figure 10 compares the predicted major maintenance schedules for 12 different bridge decks based on the developed performance prediction model, respectively, with and without concrete sealer applied every 2 years. The major maintenance activities include deck overlays and deck replacement. It is noted that some bridge decks have overlays applied more than twice at stage 4 to maintain the acceptable deck condition in a 100-year period. The results clearly show that the benefit of applying concrete sealers on the bridge deck is more significant for deck replacement and the 2nd deck overlay.

Life cycle costs of bridge decks are calculated for the service life of 100 years with a monetary discount rate of 2%. Figure 11 presents the LCCA results for comparing the life cycle costs of 12 bridge decks with and without concrete sealer in a 100-year lifespan. Based on the bridge database, it is possible that concrete sealers are applied every 2 or 4 years. Another previous study has found that concrete sealer was still effective after 12 years of application [25]. Thus, the interval of concrete sealer is selected as 2, 4, and 12 years in the analysis.

The results show that the life cycle cost of bridge decks without concrete sealers is initially lower than those with concrete sealers due to the additional costs of concrete sealers. Since deck overlays and replacement are applied earlier for the bridge decks without concrete sealer, the life cycle costs increase quickly during the period from 45th to 50th years. The concrete sealer causes similar life-cycle costs when it is applied every 2 years. As the application intervals of concrete sealers increase, the life-cycle cost is further reduced. The accumulated life cycle cost after 100 years of service life decreased from around $224/SF to $211/SF (i.e., a 6% reduction) on average when concrete sealers are applied every 12 years.

The monetary discount rate is important in LCCA to discount future costs for long-term projects. The equivalent uniform annual costs (EUACs) are calculated for all LCCA cases using different discount rates based on the range in 2023 Discount Rates for OMB Circular No. A-94. Figure 12 compares the EUACs of bridge decks with and without concrete sealers in different scenarios. The results show that the EUAC increases with the increase in discount rate. When the concrete sealer is applied every 2 years, the EUACs at all discount rates are close to those without concrete sealers. When concrete sealers are applied every 4 or 12 years, lower EUACs are observed, and the difference in EUACs between the bridge decks with and without concrete sealer is enlarged as the discount rate increases.

5. Conclusions

This study evaluated the impact of concrete sealer and salt usage on concrete bridge deck condition rating using comprehensive data from the bridge performance management system and winter maintenance records. The following conclusions are drawn from the analysis:

• The deck maintenance history shows that the average bridge age at deck overlay and deck replacement is around 25 and 50 years, respectively. Deck overlay can improve deck condition from an average rating of 6.3 to 7.1, and deck replacement can efficiently recover deck condition from an average rating of 5.3 to 8.5.
• Random Forest models were developed to predict bridge deck ratings at different stages separated by deck overlays and reconstruction. All the models show good accuracy with R-square values of 0.72–0.93. Bridge age and deck age presented the highest importance among all factors, followed by average daily traffic. The importance of the three types of salt usage is similar.
• The effect of concrete sealer on bridge deck condition is only observable at the stage before the first overlay. This indicates that concrete sealer may not be effective over the long term. More usages of prewet salt and salt brine result in slightly higher deck condition ratings, while more dry salt usage presents slightly lower deck condition ratings, indicating the benefits of salt brine over dry salt.
• Orthogonal analysis was conducted to further quantify the influence of concrete sealer and salt usage. When concrete sealer is applied every 2 or 4 years, results indicate that applying concrete sealers can help extend deck life by 1~2 years. Similarly, more usage of salt brine can extend service life by around 1~2 years.
• LCCA results show that the cost of concrete sealer cannot be neglected in analyzing its cost–benefit. The life cycle cost is similar when concrete sealer is applied every 2 years. If concrete sealer is applied every 12 years, a 6% reduction in life cycle cost could be achieved.

It is noted that the salt usage data used in this study are based on the county, not the specific bridge. This data aggregation may not reflect the exact impact of salt usage on concrete deck condition rating. Future research is recommended to validate the research findings using the refined dataset.