Prediction of Bridge Damage Factors Using Decision Tree Technique (Case Study: Concrete Rad Bridges in ZANJAN Province-IRAN)

doi:10.21203/rs.3.rs-3790975/v1

Download PDF

Research Article

Prediction of Bridge Damage Factors Using Decision Tree Technique (Case Study: Concrete Rad Bridges in ZANJAN Province-IRAN)

https://doi.org/10.21203/rs.3.rs-3790975/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

The current management system for concrete road bridges in Iran is primarily based on a descriptive model, which assesses the condition of bridge elements and assigns grades based on the level of deterioration. However, there have been concerns raised about the use of descriptive statistics when analyzing large amounts of data. In order to address these concerns, this research incorporates the data mining method known as decision tree analysis.

The decision tree analysis involves in-depth examination of data, including objective observations by experts and calculations of damage indices. We have introduced this method to replace descriptive analysis in his research. To conduct this study, various data such as physical characteristics, atmospheric and climatic conditions, traffic conditions, and deterioration indices (derived from established formulas and tables) of concrete bridges with spans exceeding 6 meters on a main 5-axis road in Zanjan province were collected. The decision tree method, along with Rapid Miner software, was utilized to model and analyze this data.

The research findings indicate that the decision tree method demonstrates a high level of accuracy in predicting and prioritizing factors that contribute to bridge damage, surpassing the capabilities of conventional models. Specifically, the study reveals that the percentage of heavy vehicles passing over a bridge is the most significant factor influencing its damage. Furthermore, it predicts that bridges with a percentage exceeding 19.5% would fall into the category of heavily damaged bridges.

Concrete Road Bridges

Data Mining

Decision Tree

Zanjan Province

forecast

In today's world, every achievement comes with its advantages and disadvantages, and human structures are no exception. One of the main shortcomings of these structures is their deterioration over time due to external factors. Bridges, being an integral part of transportation infrastructure, are particularly susceptible to such deterioration. Therefore, it is crucial to explore ways to repair and maintain them.

During the construction of concrete bridges, various human achievements are utilized, but they are not without flaws. Bridges are considered expensive infrastructure within a road network. Considering the number and extent of bridges present, it can be argued that they are the most important and costly components of the network. Therefore, achieving economic savings in bridge maintenance becomes a vital concern, along with the long-term preservation of their load-bearing capacity.

The goals and programs associated with bridge maintenance include:

Maintaining the usability and load-bearing capacity of bridges for as long as possible.
Achieving economic savings in both the present and future.
Ensuring continuous operation within existing financial constraints.
Guaranteeing the safety of passengers using the bridges.
Minimizing disruptions to through traffic caused by bridge maintenance activities.
Establishing enough confidence among travelers to provide a comfortable journey experience.

These goals reflect the importance of maintaining bridges in a cost-effective manner while ensuring their functionality, safety, and convenience for users.

The Bridge Management System (BMS) is a management system utilized by organizations responsible for the construction and maintenance of bridges. Its purpose is to maintain the efficiency and service level of bridges while effectively managing maintenance costs. The Iran Road Maintenance and Transportation Organization has recognized the importance of BMS and has incorporated it into their agenda since 2018, aligning with their goals for excellence.

Despite the features offered by BMS systems used worldwide, bridge managers often struggle to identify high-risk situations that bridges may encounter throughout their lifespan. This difficulty arises due to several factors, including the large number of bridges to manage (approximately 350,000 in Iran) and the multitude of physical, climatic, and traffic-related factors that impact bridge conditions. As a result, establishing a logical relationship between these influencing factors becomes challenging, and descriptive statistics alone are insufficient in providing appropriate responses to these complex conditions.

In order to address these challenges, this study employs data mining methods to analyze the vast amount of data and prioritize bridges accordingly. Data mining is utilized to extract hidden patterns and insights from the data. By applying data mining techniques, it becomes possible to identify key factors and prioritize bridges based on their risk levels.

Data mining, also known as DM, is a field that involves the extraction of hidden information from large datasets (Shoba and Shobha 2014). It is considered both an art and a science (Rajakumari and Nalini 2016). The main objective of data mining is to discover new, unexpected, and valuable patterns in data. This process can be applied to various sources of data, such as text, databases, transaction data, multimedia data, sequences, web data, time series, spatio-temporal data, graphs, social networks, and informational data (Krishnamoorthy and Karthikeyan 2022).

The need for data mining arises from the increasing size of databases, which makes it difficult for humans to comprehend and extract meaningful insights from them. Without powerful tools, decision-makers are unable to extract valuable information from the abundance of data available. This situation highlights the challenge of having a wealth of data but a lack of actionable information. In today's highly competitive environment, using data effectively has become a crucial goal for managers to improve the current state of affairs. (Shahrabi 2013)

The main methods of data mining can be categorized into two groups: Descriptive and Predictive. The Descriptive method aims to determine the general properties of data and find human-interpretable patterns. On the other hand, the Predictive method focuses on predicting future behavior by utilizing multiple variables or fields in the database to estimate unknown values of other variables.

Classification is a key function of the Predictive method, involving the process of finding a model that can predict unknown categories of objects by identifying data categories or concepts. Essentially, classification is a learning function that assigns a data item to one of the predefined categories (Lyu et al. 2010). One commonly used method for classification is the Decision Tree or Classification and Regression Tree (CART). A Decision tree is an algorithmic representation of branching that resembles a flowchart. The root of the tree represents the highest node, with each branch representing a test output and the leaf nodes representing the distribution of categories (Soleimanpour et al. 2016).

Decision trees are employed to predict the membership of objects into different categories. The flexibility of this technique has made it a popular choice in data mining. The decision tree methodology consists of two main phases: (Cichosz 2014)

A. Creating the primary tree: The decision tree is constructed using the training data set, continuing until each leaf becomes pure (homogeneous).

B. Pruning: The grown tree is pruned in this phase, based on the experimental data set, to enhance the accuracy of the model.

In this research, the literature review section provides an overview of data mining techniques, as well as the bridge management system and its local model. Additionally, several previous studies that have utilized the decision tree technique in the field of bridge and infrastructure management are discussed. To investigate the impact of different factors on the damage of concrete bridges and their ranking, a decision tree model was constructed. Bridge data, including physical characteristics, climatic conditions, traffic conditions, and structural damage indexes, were extracted for bridges with a span above 5 meters on the main axis. The data was then normalized after necessary corrections.

The analysis of this data was performed using Rapid Miner software, which facilitated the creation of the decision tree model. By examining the model output, the factors that have the highest impact on the damage index were identified, and conditions leading to critical damage modes were predicted. Finally, the research concludes with a presentation of findings and recommendations based on the study's outcomes.

Data Mining (DM) & Decision Tree (CART)

Data mining is an iterative process rather than a linear one. Its main objectives are to describe, predict, and prescribe (Dominguez et al. 2023). The concept of classification and regression trees was introduced by Braiman. This method aims to divide data into smaller sections to maximize the separation of information (Gordon 2013). The decision tree algorithm is a predictive modeling technique that recursively divides input data and can handle predictor variables as well as numerical and categorical targets (Loh 2011). In essence, a decision tree is a symbolic learning method that organizes information from a training data set into a hierarchical structure with nodes and branches. All decision tree algorithms define two types of variables: independent variables and dependent variables. The dependent variable, or target variable, takes discrete values in a classification tree and continuous values in a regression tree. To create a decision tree, the data set is divided into training data and test data. The decision tree is built based on the training data, and then the trees are pruned to prevent overfitting. Two examples of common decision tree algorithms are CART C and 5.0 (Nie et al. 2011).

To use the CART algorithm, a target attribute in the dataset must be specified. The algorithm starts from the root node and divides the training data into two child nodes, which are further divided into grandchild nodes. During tree growth, the CART algorithm searches for questions that separate each parent node into two homogeneous child nodes. Homogeneity means that all records in a node belong to a specific category, in which case the node becomes a leaf. The algorithm looks for features in the dataset that have strong separating properties. As the tree grows, the nodes become more homogeneous and display more information. Tree growth continues until the tree reaches its maximum size and the division operation stops due to lack of data (Sharma and Kumar 2016).

Bridge Management System (BMS)

he primary objective of bridge management is to safeguard the significant assets utilized in bridges while minimizing operational costs. Bridge management encompasses time-consistent economic planning, preventive and corrective maintenance, and even replacement of bridges with minimal disruption to traffic. This entails systematic inspections, specialized evaluations, and effective information and program implementation (Ryall et al. 2000).

Implementing a bridge management system allows managers to organize assets hierarchically based on their condition, importance, and maintenance or repair requirements. By adopting a BMS, asset managers gain a suitable framework for prioritizing investments from both technical and economic perspectives (Alonso and González 2022). A bridge management system is designed to optimize the utilization of available resources for inspecting, maintaining, reconstructing, repairing, and replacing bridges. The main purpose of developing a BMS is to equip maintenance offices with the capability to identify the general condition of each bridge in the road network at different intervals, as well as the condition of specific bridge components. Additionally, a BMS should employ analytical or statistical methods to determine possible maintenance and reconstruction programs, along with the associated costs (Tonias 1995).

Achieving these objectives is undoubtedly challenging. The creation and development of a BMS are complicated by the size and complexity of the database that assesses the structural condition and the evolving nature of structures in terms of design systems and materials used. Without an effective bridge management system, decision-making related to bridge inspections and maintenance becomes fragmented, and budgets are allocated without considering the overall network of bridges and its impact (Tonias 1995).

BMS Modules

Since the Bridge Management System (BMS) is a highly flexible and reliable modular system, it can easily be customized to meet the specific routines and special requirements of different organizations in various countries. The BMS generally consists of nine modules, as outlined below (COWI 2004):

Registration of Information Forms: This module involves the technical identification of bridges and the collection of relevant information.
Initial (Superficial) Inspection: This module focuses on conducting preliminary inspections to assess the general condition of bridges.
Main Inspection and Assessment of Conditions: This module is dedicated to performing comprehensive inspections to evaluate the overall condition of bridges.
Special Inspection: This module is designed for conducting specialized inspections to address specific concerns or issues related to bridges.
Load-Bearing Capacity and Vehicle Evaluation: This module involves assessing the load-bearing capacity of bridges and evaluating their compatibility with different types of vehicles.
Preparation of a Maintenance Strategy: In this module, a maintenance strategy is developed to ensure the proper upkeep and preservation of bridges.
Prioritization of Projects from a Technical and Economic Perspective: This module focuses on prioritizing bridge projects based on technical and economic considerations.
Budgeting Maintenance Works: This module involves budgeting and allocating resources for bridge maintenance activities.
Transportation Management of Heavy and Super Heavy Traffic Loads: This module is dedicated to managing the transportation of heavy and super heavy traffic loads, taking into account the structural integrity of bridges.

Overall, BMS offers a comprehensive solution for effectively managing bridges, catering to the specific needs of each organization and country.

Bridge Management System Method Used in Iran

The local version of the road bridge management system in the country has been a priority for the Iran Road Maintenance and Transportation Organization since 2018. The implementation of this system involves two stages. In the first stage, the bridge information form is completed and uploaded into a web-based software that has been developed for this purpose. This form includes essential details about the bridges in the system. In the second stage, the system allows for the registration of bridge damages and related information. Users can upload images and additional details about the damages into the system.

The system then utilizes a scoring method (unweighted) to determine the priority of bridges. This scoring is based on factors such as the overall damage index of the bridge structures, the damage index during bridge operation, and the general condition ranking of the bridge. By implementing this local version of the road bridge management system, the Iran Road Maintenance and Transportation Organization aims to improve the efficiency and effectiveness of bridge maintenance and prioritize necessary repairs based on the condition and impact of damages. The models utilized in this version are as follows (Fig. 1):

a. ID card model

b. Administrative data: organization (3 categories)

c. Technical data: geometric data, global coordinates, spans, design (14 categories)

d. Main inspection module: conducting the main inspection, conducting the overview, recording the damages, the responsible checklist, evaluating the members, and preparing the report (Scoring the damage index of the entire bridge structures, damage index of the bridge operation, and ranking the overall condition of the bridge)

In this model, the structural indicators of bridges have been determined based on the information collected through the objective observation method. To calculate these indicators, a list of bridge damages has been compiled and coefficients have been assigned to each damage type. The structural index is obtained by summing the values of all the bridge members and multiplying them by their respective coefficients, which take into account factors such as the type and amount of damage, length and height of the bridge, and the complexity of the cross-sectional design. Additionally, an operation index is calculated for damages related to drainage, surface, protection, fence, sidewalk, and cornice. The structural index is calculated separately for other bridge members.

Similar research in bridge management and decision tree algorithms(Table1)

In his paper, Morcous (2005) presents a model using the decision tree algorithm to predict the impact of factors such as average daily traffic, truck percentages, and environmental impact on bridge deck damage.
Li et al. (2005) utilized the decision tree method to predict the effect of tornadoes on the vibration of suspension bridges. They selected wind speed and wind direction in three different locations along the bridge span as model inputs.
Lu et al. (2005) employed the decision tree algorithm to predict the condition of bridge components. They demonstrated that bridge deck conditions are influenced by factors such as the type of deck structure, bridge age, bridge road width and base highway network, pavement conditions with road improvement cost, road width and inspection frequency, and infrastructure conditions with road defect cost and deck width.
Yau et al. (2005) investigated big data analysis methods, including neural networks, decision tree, and genetic algorithm, to predict possible breakdowns in Taiwan's bridges for maintenance purposes. The study focused on a 15-year-old dataset.
Kameshwar et al. (2005) proposed a model for bridges restoration using the decision tree algorithm to evaluate extreme events and crises in road networks. The method involved obtaining seismic repair models of bridges and utilizing three separate decision trees to determine different traffic constraints predicted for varying levels of damage to bridge components, including bridge closure, line closure, and speed/load limits.

Table 1

Summary of data mining studies
Item	Authors	Technique	Data Mining Field
1	Morcous	Decision Tree Algorithm	Predicting the effect of average daily traffic, trucks percentage and environmental effects on bridge deck damage
2	Li et al.	Decision Tree Algorithm	Predicting the effect of tornadoes on the vibration of suspension bridges
3	Lu et al.	Decision Tree Algorithm	Predicting the condition of bridge components
4	Yau et al.	Neural networks, decision tree and genetic algorithm	Predicting possible damages to Taiwan's bridges
5	Kameshwar et al.	Decision Tree Algorithm	Seismic repair modeling of bridges in extreme events

Research Objectives

1. Predicting the impact of physical, climatic and traffic factors on bridge damage

2. Prioritizing the effective factors in the critical damages of bridges

Statistical Population

The statistical population of this research includes all concrete road bridges of 5 main axes of Zanjan province with a length of above 5 meters, including 384 bridges.

Data collection method

In this study, the researchers utilized the databases of Zanjan Road Maintenance and Transportation Organization, as well as the statistical yearbook of the Meteorological Organization for the year 2019. These sources were selected due to their relevance to the study's type, purpose, subject, and the characteristics of the statistical population under investigation. The information gathered from these databases played a crucial role in conducting the research.

Model inputs and outputs

In the previous sections, it was mentioned that the measurement of bridge damage is based on the Iranian model, which incorporates structural damage and operation indicators. These indicators are obtained through the observations of inspectors and calculations, serving as the variables in the model. When selecting model inputs, the findings of other researchers such as Morcous, Li, and Lu, who have studied factors influencing the damage of concrete bridges, have been utilized. Naturally, among the numerous factors, only those for which data could be collected and validated in the statistical population have been considered.

Consequently, three sets of data can be employed to determine the efficiency of bridges based on the output of structural damage and operation indicators. These three groups consist of climatic factors (e.g., number and amount of rainy and snowy days, number of frosty days, average daily temperature difference, intensity and acidity of rain and snow, salt concentration used in snow clearance and de-icing operations), traffic factors (e.g., volume of traffic, heavy vehicle traffic, tonnage crossing the bridge, speed of crossing the bridge, number of vehicle collisions with the bridge deck and supports), and maintenance factors (e.g., frequency and types of cross-sectional and major repairs, maintenance cost). Additionally, physical factors such as bridge length, largest span length, bridge width, and bridge height are taken into account.

Unfortunately, there was insufficient information regarding accidents that occurred on bridges, as well as maintenance records and related costs in the provincial road maintenance and transportation organizations. As a result, data from the provincial road accident registration system (Provincial Roads Management Center), intelligent road traffic system, and meteorological systems in the area of bridges, along with important physical characteristics affecting structural damage, were extracted and used as model data after validation and correction. In the following section, we will examine the inputs and outputs utilized in the model. (Table 2)

a. Model input

Meteorological and Climatic Information:

(Sources: Reports from the General Directorate of Meteorology of Zanjan Province and the Road Incident Registration System of the Road Maintenance and Transportation Organization of Iran)

Input 1: Number of Axial Snowfalls

- Definition: The number of times snowfall has been reported on the mentioned axis section during the year, based on reports from the Road Management Center.

Input 2: Number of Axial Rains

- Definition: The number of times rain has been reported on the mentioned axis section during the year, based on reports from the Road Management Center.

Input 3: Amount of Rainfall (ml)

- Definition: The total amount of rain and snowfall on the mentioned axis section during the year, according to reports from the Road Management Center.

Input 4: Number of Frosty Days

- Definition: The number of days with frost in the bridge area, based on reports from meteorological stations near the bridge.

Input 5: Number of Days of Snowfall

- Definition: The number of days with snowfall in the bridge area, according to reports from meteorological stations near the bridge.

Input 6: Average Snowfall/Amount of Rainfall (ml)

- Definition: The average snowfall or rainfall in the bridge area, based on reports from meteorological stations near the bridge.

Input 7: Average Temperature Difference

- Definition: The average temperature difference between day and night in the bridge area, according to reports from meteorological stations near the bridge.

Traffic Information:

(Source: Traffic Counting System of the Road Maintenance and Transportation Organization of Iran)

Input 8: Average Daily Traffic (ADT)

- Definition: The average daily round-trip traffic on the axis, based on information from traffic counts installed on the axis.

Input 9: Percentage of Heavy Vehicles

- Definition: The percentage of heavy vehicle traffic (including trucks, trailers, and buses) in relation to the total traffic of vehicles on the axis, based on information from traffic counts installed on the axis.

Input 10: Average Speed (km/h)

- Definition: The average speed of vehicles traveling on the axis, based on information from traffic counts installed on the axis.

Physical Information of Bridges:

(Source: Bridge Management System of the Road Maintenance and Transportation Organization of Iran)

Input 11: Length of the Bridge

b. Model Outputs

(Source: Bridge Management System of the Road Maintenance and Transportation Organization of Iran)

Output 1: Bridge Structural Damage Index

- Definition: A measure of the structural damage of the bridge, determined based on the damage status of each bridge element, application of member impact coefficients, and the degree of damage.

Table 2

Model Variables
Dependent Variable		Description			Code
Dependent Variable		Structural damage index			O1
Dependent Variables	Meteorology	Description	Code	Description	Code	Description			Code
		Number of axial snowfalls	I1	Number of axial rains	I2	Amount of rainfall (ml)			I3
		Number of frosty days	I4	Number of snowfall days	I5	Average rainfalls	I6	Average temperature difference	I7
	Traffic	Average daily traffic (ADT)	I8	Heavy vehicle percentage	I9
	Physical	Bridge length	I11

Information Analysis Method

In this study, our objective is to prioritize the factors that affect the structural damage of bridges and predict their degree of importance and destructive scenarios. To achieve this, we have employed the decision tree method. This method allows us to select the independent variables, which include the physical characteristics of the bridges as well as atmospheric, climatic, and traffic conditions. The dependent variables are represented by structural damage indicators based on the Iranian BMS model.

Before entering the model, we performed data preparation operations to ensure that the data is suitable for analysis in Rapid Miner 5.1 modeling software. It is important to note that any errors or inaccuracies in the data preparation process can lead to incorrect outputs or inaccurate predictions. The changes made to the model are presented in Table 3, while Fig. 2 depicts the model created in the software.

To evaluate the model's performance, we selected 10% of the statistical population data as test data using the sample command. The remaining data was used as training data, with missing values being substituted using an appropriate method. Data preparation was conducted using Excel software before loading it into the model.

In this model, we discretized the dependent variable of the training data based on frequency. Discretization involves converting numerical data into nominal data by dividing the numerical values of a particular attribute into several intervals.

Table 3

Characteristics of model variables
Role	Name	Type	Statistics	Range
prediction	code	polynominal	Mode = a523 (1), least = a4 (0)	a109(1), a147 (1), a170(1), a17
regular	Structural damage index	numeric	avg = 22.894 +/- 24.573	[0.000 ; 97.200]
regular	Bridge length	numeric	avg = 13.634 +/- 21.661	[5.000 ; 136.000]
regular	Axial snow	integer	avg = 9.974 +/- 5.504	[2.000 ; 19.000]
regular	Axial rain	integer	avg = 11.579 +/- 7.195	[2.000 ; 25.000]
regular	Rainfall	integer	avg = 357.053 +/- 46.303	[277.000 ; 392.000]
regular	Number of frosty days	real	avg = 103.605 +/- 14.444	[80.000 ; 116.000]
regular	Number of snowfall days	real	avg = 35.500 +/- 7.059	[26.000 ; 43.000]
regular	Rainfall average	integer	avg = 0.979 +/- 0.128	[0.760 ; 1.070]
regular	Average temperature difference	numeric	avg = 13.797 +/- 1.032	[12.400 ; 14.800]

Data Analysis

There are several methods available for implementing and executing data mining projects, and one of the most powerful methodologies is the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology. In this paper, we present a proposed CRISP-based model that comprises five phases, each consisting of sub-sections. It is important to note that moving forward and backward between different phases is necessary, as the input for each phase depends on the output of the previous phase (Al Jarullah 2011). In the following sections, we will examine each phase of this proposed model in accordance with the case study.

a. System Recognition

During this phase, the first step involves recognizing the desired system. Subsequently, the goals and key success factors of the system are determined and reviewed. Table 4 provides an overview of the model variables and their properties. The code variable is defined in the I.D. role, while the structural damage index variable is defined in the prediction role. Since there is no missing data in this model, there is no need for any data modifications.

b. Data Collection and Recognition:

In this phase, raw data was collected from Iran's bridge management system reports. The collected data was then validated and normalized using M.S. EXCEL software. Table 4 presents a portion of the raw bridge information prior to the preparation process.

Table 4

Part of the raw information of bridges
Code	Bridge length	Axial snow	Axial rain	rainfall	Number of frosty days	Number of snowfall days	Rainfall average	Average temperature difference	Average daily traffic (ADT)	Heavy vehicle percentage
a1	16	10	24	385	116	43	1.06	14.8	2181	21.5
a10	10	3	3	277	80	28	0.76	13.1	1729	39
a100	14	3	3	277	80	28	0.76	13.1	1729	39
a101	6	10	23	385	116	43	1.06	14.8	2181	21.5
a102	85	6	12	316	112	26	0.87	13.2	12137	31.42
a103	10	3	3	277	80	28	0.76	13.1	1729	39
a104	7	3	3	277	80	28	0.76	13.1	1729	39
a105	8	2	3	295	110	34	0.8	14.3	7506	9.5
a106	12	16	17	392	91	30	1.07	12.4	15736	24.63
a107	12	2	2	277	80	28	0.76	13.1	1729	39
a108	6	3	2	277	80	28	0.76	13.1	1729	39
a109	5	12	8	392	91	30	1.07	12.4	2536	27
a11	44	6	5	385	116	43	1.06	14.8	2536	27
a110	9	10	23	385	116	43	1.06	14.8	2181	21.5
a111	6	3	4	277	80	28	0.76	13.1	1729	39
a112	13	3	3	277	80	28	0.76	13.1	1729	39
a113	13	3	5	295	110	34	0.8	14.3	7506	9.5

c. Modelling

The methodology used in this study is predictive data mining, specifically employing the decision tree algorithm to establish relationships between different variables. In this stage, the C5.0 decision tree algorithm was tested with various inputs. A decision tree provides a clear representation of branching using an algorithm. Each leaf node in the tree represents a subset of the training data, with each instance belonging to a specific leaf node. (Witten and Frank 2005)

The modeling was conducted using Rapid Miner 5.2 software. The model consists of two branches: "entering and assigning roles to training data and setting analysis components" and "entering test data" (Fig. 2). The structural damage index, which serves as the prediction data, is divided into five ranges proposed by the software. Domain 5 represents bridges in very poor condition, while domain 1 indicates bridges in a safe condition with minor damage (Table 4).

The Decision Tree generated by the software determines the range of the structural damage index based on different values of the independent variables (Fig. 3). According to this tree, the worst scenario for bridge damage, which corresponds to Range 3 (damage index ranging from 11.976 to 167.962), occurs under the following conditions:

The axis experiences more than 13.5 days of snowfall per year, the average daily traffic is less than 13,936 vehicles, and the percentage of heavy vehicles in the total traffic exceeds 15.5%.
The axis experiences more than 23 days of rainfall per year, more than 7 days of snowfall per year, the average traffic is less than 2,358 vehicles per day, the region has more than 27 snowy days per year, the annual rainfall exceeds 388 mm, and the percentage of heavy vehicle traffic is more than 19.5% of the total traffic.

Different scenarios can be derived from the decision tree for the high (Range 2) and medium (Range 1) damage index ranges as well. One important observation is the inverse relationship between the daily traffic index and the structural damage index. This contradicts the commonly held belief among bridge experts that higher traffic volume and weight lead to increased fatigue and damage to the bridge structure over its lifespan. This contradiction could be attributed to issues with the formulas or model tables used, or it could indicate errors in the observation and diagnosis of damages or recording in the software.

Furthermore, the decision tree presented does not mention independent variables such as freezing days, average rainfall, rainfall in the axis, temperature difference, and bridge length. This suggests that either these variables have little or no significant relationship with the structural damage index, or they are considered less important by the software compared to other variables and have been excluded from the decision tree. In other words, variables such as snowfall and heavy rain amounts, the percentage of heavy traffic compared to total traffic, the number of snowy days, and the daily traffic volume have a more significant impact on bridge damage, particularly the percentage of heavy vehicles in relation to total traffic.

d. Evaluation

In this phase, it is crucial to evaluate the results of the modeling process to ensure the model's effectiveness and usability. To assess the accuracy of the model, the available data is divided into three parts: training, testing, and validation. The training data is used to build the model, while the testing data is employed to evaluate the model's performance by determining the labels for the test records. The validation data further verifies the accuracy of the model.

In this particular model, 10% of the statistical population data was assigned as the test data using the sample command, while the remaining data was used as the training data with the replacement method. Prior to entering the model, data preparation was conducted using Excel software and then loaded into the model. Additionally, the dependent variable of educational data underwent discretization based on frequency, which involves converting numerical data into nominal data by dividing the numerical values into intervals.

The accuracy and precision of the model on the training data set are measured as the percentage of correctly classified observations. This is calculated using the test data and the error rate. The misclassification rate can also be derived from the accuracy index. In order to calculate these indices, several definitions need to be considered. "True positive" refers to positive data that is correctly predicted by the algorithm, while "true negative" represents negative data that is correctly predicted. "False positive" indicates negative data that is incorrectly predicted, and "negative errors" are positive data that are inaccurately predicted. (Han and Kamber 2006)

For evaluating the algorithm's performance in this paper, the precision index is utilized. This index signifies the correct prediction rate of the classification algorithm. Based on the results of the model (Fig. 4), the model exhibits an error rate of 3.66% and an accuracy of 96.38%. Therefore, the model's validity can be considered acceptable.

e. Development

Creating a model is not the end of a project; the purpose of data mining projects is to discover knowledge and utilize the knowledge in the future. The discovered knowledge needs to be organized and made accessible to others. In the first step, we utilized the Apply Model operator in Rapid Miner software and the Gini-index method to predict the condition of the structural damage index of bridges based on data mining performed on the entire dataset (Fig. 2).

Given that the model focuses on the special feature of the structural damage index, which has 5 ranges based on the tested bridges' damage indexes, Fig. 5 shows the software's display of the probability for each bridge to fall within these 5 ranges and predicts the most likely range. For example, for the bridge with code a23, the software predicts a 94% probability for it to be in damage index range 1 and a 6% probability for range 2.

Another advantage of Rapid Miner software in developing data mining models is the creation of scatter plots. These plots provide valuable insights for researchers to study the distribution of bridges based on the factors that influence the structural damage index. In Fig. 6, the scatter plot shows that bridges in the middle 1 range (approximately 25) have a percentage of heavy vehicles around 30%. This figure helps identify situations where certain indicators can increase the risk of damage.

In relation to the aforementioned topics, it can be argued that the effective utilization and extraction of hidden data from the vast amount of data generated from the maintenance and repair processes of numerous bridges over several years is an undeniable necessity. This research focuses on examining the feasibility of a tree model in analyzing this data volume and confirming its decision-making capabilities. By employing data mining techniques as an operational tool, the classification and ranking of bridges can be achieved, thereby addressing the major challenges and shortcomings of descriptive bridge management models. Moreover, the predictive capabilities enabled by this approach allow for the identification of potential scenarios that may lead to severe damage or even destruction of bridges during their lifespan.

Consequently, the proposed method was applied to analyze data related to road bridges in Zanjan province. The model, constructed using Rapid Mainer software, utilized the loading and structural damage index as a distinguishing characteristic for selecting and calculating experimental data based on training data. The model demonstrated a high validity rate of 96.38% and indicated that the most significant factor contributing to bridge damage, as identified by the decision tree, was the percentage of heavy vehicles in relation to the total traffic crossing the bridge.

Furthermore, the analysis of the decision tree resulting from this project revealed the worst-case scenario for bridge damage, outlined in Table 5. These conditions represent situations where the risk of damage is expected to increase, and therefore should be prioritized in terms of maintenance efforts.

Table 5

The worst damage scenario
range of structural damage	Quantity	The effective index	Code
big damage	More than 19.5	Heavy vehicle percentage	I9
big damage	More than 13.5	Number of axial snowfalls	I1
Moderate damage	More than 24	Number of axial rains	I2
big damage	More than 27	Number of snowfall days	I5

Another result of data mining is the identification of the dispersion state of bridges in different ranges of indicators that affect bridge damage. For example, in Zanjan province, concrete road bridges with a structural damage range of 0–55 experience nearly a 30% increase in passing traffic loads compared to their normal loads. Another significant finding from this research is the exclusion of certain independent variables, which typically have a direct relationship with the structural damage index, from the formulation of bridge damage scenarios. This issue raises concerns about the existing model for managing suburban bridges in Iran. The current model relies solely on visual observations by experts, scoring elements and their damage rates using tables to determine the damage index. It lacks a mechanism for data validation and lacks the use of analytical methods when dealing with the large amount of bridge data, which questions the accuracy of the model.

Ultimately, the identification of different damage scenarios can assist bridge maintenance and repair authorities in considering the threshold of physical, climatic, and traffic factors that increase the risk of damage. This information can be used as inspection and budget criteria, aiding in the classification and crisis management decisions made by provincial management authorities.

Based on the intriguing findings of this study, it is recommended that future research focuses on developing a model to accurately predict the effects of climatic, meteorological, and traffic factors on the damage of different bridge elements. This will help identify the most vulnerable elements and their sensitivity to these factors.

Author Contribution

Ali medghalchi wrote the main manuscript text and All authors reviewed the manuscript.

Acknowledgment:

I would like to express my gratitude to all the maintenance experts at Zanjan Road Maintenance and Transportation Organization, particularly Mr. Hossein Eskandari and Mr. Naderloo, for their valuable cooperation in creating the database of Zanjan bridges.

Shoba, G. and G. Shobha 2014."Water Quality prediction using data mining techniques: A survey." International Journal of Engineering and Computer Science, Vol. 3 No. 06.
Rajakumari, S.B. and C. Nalini.2016."Identification of lead contaminant in river water quality data." Journal of Chemical and Pharmaceutical Sciences.
Krishnamoorthy, M. and R. Karthikeyan.2022. "Pattern mining algorithms for data streams using itemset." Measurement: Sensors,Vol 24, 100421.
Shahrabi, J. 2013."Data Mining." Tehran: Jahad-e Daneshghahi Publication (in Persian).
Lyu, S-R., W.T. Wu, C.C. Hou and W-H. Hsieh. 2010. "Study of cryopreservation of particular chondrocytes using the Taguchi method.", Cryobiology, Vol. 60: 165–176.
Soleimanpour, S., S. Mesbah and B. Hedayati.2016."The applicationof data mining algorithm K-Means and CART most influential factors in determining the quality of drinking water in Nurabad plain of Fars Province." Proceedings of Eleventh National Congress of Watershed Management Science and Engineering, Yasouj, Iran. (in Persian)
Cichosz, P. 2014."Data Mining Algorithms: Explained Using R. New York." John Wiley & Sons.
Dominguez, X., A. Prado, P. Arboleya, and V. Terzija.2023. "Evolution of knowledge mining from data in power systems: The Big Data Analytics breakthrough."Electric Power Systems Research, Vol 218, 109193.
Gordon, L. 2013."Using classification and regression trees (CART) in SAS® enterprise miner TM for applications in public health." Public Health, Paper 089.
Loh, WY.2011. "Classification and regression trees." WileyInterdisciplinary Reviews: Data Mining and KnowledgeDiscovery, https://doi.org/10.1002/widm.8.
Nie, G, W. Rowe, L. Zhang, Y. Tian and Y. Shi 2011. "Credit card churn forecasting by logistic regression and decision tree." Expert Systems with Applications, Vol 38, Issue 12.
Sharma, H. and S. Kumar 2016." A survey on decision tree algorithms of classification in data mining." International Journal of Science and Research, Volume 5, Issue 4.
Ryall, M.J. and G.A.R. Parke and J.E. Harding 2000. "Bridge Management, Inspection, Maintenance, Assessment & Repair." Springer, Uk.
Alonso, M.P. and J.L. González 2022." Reinforced concrete long-term deterioration prediction for the implementation of a Bridge Management System."materials today proceedings, Vol 58, Part 4.
Tonias, D.E. 1995. "Bridge Engineering, Design, Rehabilitation and Maintenance of Modern Highway Bridges." McGraw-Hill.
COWI, 2004." BRIDGE OPERATION & MAINTENANC." Technical Report.
Morcous, G. 2005. "Modeling Bridge Deck Deterioration by Using Decision Tree Algorithms." Transportation Research Record: Journal of the Transportation Research Board, DOI:10.3141/trr.11s.e383j231l68k41h2.
Li, S., S. Laima and H. Li 2018. "Data-driven modeling of vortex-induced vibration of a long-span suspension bridge using decision tree learning and support vector regression." Journal of Wind Engineering and Industrial Aerodynamics, Vol 172.
Lu, P. and Z. Zijian 2017. "Bridge Components Condition Rating Prediction Using Decision Tree Model, Global Environment." Health and Safety, vol1.
Yau, N. and Y. Chuang 2015. "Analyzing Taiwan Bridge Management System for decision making in bridge maintenance: A big data approach." 10th International Joint Conference on Software Technologies (ICSOFT), Colmar, France.
Kameshwar, S., S. Misra and J. Padgett 2020. "Decision tree based bridge restoration models for extreme event performance assessment of regional road networks." Journal Structure and Infrastructure Engineering, Vol 16.
Al Jarullah, A. 2011. "Decision tree discovery forthe diagnosis of type II diabetes." International Conference on Innovations in Information Technology.
Witten, I.H. and E. Frank 2005." Data Mining: Practical machine learning tools and techniques." Morgan Kaufmann.
Han, J. and M. Kamber 2006."chapter 1: introduction: Data Mining: Concepts and Techniques." Morgan Kaufman Publisher. 2nd ed.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Prediction of Bridge Damage Factors Using Decision Tree Technique (Case Study: Concrete Rad Bridges in ZANJAN Province-IRAN)

Status:

Version 1

Abstract

Figures

Introduction

Literature Review

Research Methodology

Conclusion and Recommendations

Recommendations

Declarations

Author Contribution

Acknowledgment:

References

Additional Declarations

Status:

Version 1