Performance Evaluation of Low-Volume Flexible Pavement

Abstract

Low-volume rural roads (LVRRs) play a crucial role in the economic, cultural and heritage development of any nation. LVRRs are part of a built-in road system in all developed and developing countries. India has a total road network of over 6.32 million kilometers in length of roadways, making it the second largest road network in the world. The LVRRs in India constitute about 75% of all categories of roads. These roads act as a lifeline to the rural communities in terms of providing access to schools, markets, medical, recreational and commercial activities. Since many of the road agencies feel that these roads carry low traffic, a sufficient maintenance budget is not allocated to keep these roads to user satisfaction. As a result, the rate of deterioration of these pavements is higher; this can be addressed by timely maintenance. It is in this context that the performance evaluation of low-volume rural road flexible pavement and the development of a performance prediction model for in-service pavements are essential. Performance models are developed using the data collected on critical distress parameters. This study investigated the significant factors that affect pavement performance at the project and the network investigated, and models were developed using an artificial neural network and genetic algorithm for the prediction of future conditions of the network.

1. Introduction

Rural road connectivity, as the critical element for the economic development of the nation, has been accepted and proven worldwide. LVRRs, as used in India, are referred to as low-volume rural roads (LVRRs) around the globe. India started road development with a series of plans since its independence. However, no specific attention was given to rural roads. In the absence of a central authority to monitor design/specifications, states adopted their methods in road construction, including stage development with subjective decisions under budgetary constraints almost till the end of 2000. The Government of India, being convinced of the need for the scientific development of these roads, launched the “Prime Minister’s Rural Roads Program” (Pradhan Mantri Gram Sadak Yojana) PMGSY (2000) [1], with the intent of providing all-weather roads with the full association of all State Governments. The programme started with proper network planning for a realistic assessment of needs, a substantial focus on quality with a three-tier mechanism and suitable technologies in the design, construction and maintenance of roads in rural India. The programme is constantly on the lookout for new and innovative technologies while revising and updating the codes of practice with due importance to knowledge dissemination. The objectives have also been enhanced in different stages of the programme during the past two decades. Having accomplished the majority of set goals, the programme is now on a path to Asset Management (AM) to achieve sustainable rural development.

Further, a vast road network has brought connectivity to the rural areas. Earlier, these roads were constructed as gravel or earthen roads. However, these were upgraded to the engineered structure by providing a thin bituminous surfacing layer of Open Graded Premix Carpet (OGPC) of 20 mm as per the IRC: SP: 72–2015, [2] guidelines. The methods that are used for the maintenance of these roads are based on the eye judgement, ad-hocism and experience of engineers without considering actual pavement performance data; as a result, these roads are damaged after one or two monsoons. So, it is necessary to plan regular maintenance that would minimise not only the usage of financial resources but also other resources such as equipment, workforce and materials. A study conducted by [3] pointed out that the road in deplorable condition is four to five times the cost if a pavement is regularly maintained while it is in good condition. The allocation of budget towards the maintenance of low-volume rural roads is always on the lower side when compared with the high-volume highways. Further, the tendency is to neglect those roads that have low traffic volume and a low composition of light commercial vehicles. This has led to different procedures being adopted for the maintenance of LVRRs, and therefore, allocated resources should be used wisely [4].

In view of the growing importance of low-volume rural roads, the construction, maintenance and evaluation of their performance are of paramount importance. Figure 1 shows a view of a typical low-volume rural road in India. Various research institutes and agencies have attempted over the last 30 years to develop models to predict the performance models of high-volume highway pavements. However, very few studies have been carried out on LVRRs. Therefore, this chapter presents the performance models developed to describe the predicted pavement condition. Using these models, one can predict the future pavement condition and plan the maintenance and budget requirements accordingly.

Figure 1.

View of a typical low-volume road in India.

2. Study area and data collection

In this study, a total of fifteen road stretches were selected in three districts of Andhra Pradesh, India, to represent different geographical and environmental conditions. After identifying the test sections according to the above-said criteria, the performance evaluation of in-service pavements was conducted for 3 years during pre-monsoon and post-monsoon. The design details and historical data on test sections were compiled from the Detailed Project Reports (DPR) of the project roads. After identifying the test sections, the road inventory (Height of embankment, width of carriageway, width of shoulders and side slopes) details of the pavement sections were measured in the first round of investigation. The pavement evaluation studies were conducted over 3 years. Pavement distress due to structural failure and functional failure in terms of rutting and roughness, respectively, were measured during the same periods. MERLIN equipment was used to collect the roughness data at regular intervals along the road. The structural evaluation was done using a Portable Falling Weight Deflectometer (PFWD) and dynamic cone penetrometer. The research approach for the development of performance models is illustrated in Figure 2.

Figure 2.

Research approach.

3. Research approach

One of the critical tasks of this study was to investigate the significant factors that affect the pavement performance of LVRRs at project and network levels. To achieve this target, the trend between different independent parameters was plotted to examine the behaviour and interactions of different parameters considered. This has given an idea of the selection of influencing parameters for the model-building process. Taking this into account, initially, models were developed for each road level, and those models were integrated to develop project and network models for each distress. An iterative linear and non-linear regression analysis was adopted to determine the model coefficients as well as artificial neural network (ANN) and genetic algorithms techniques.

3.1 Modelling approach for pavement performance

One of the critical goals of this research was to investigate the significant factors that affect pavement performance at the project and network levels. Every likely variable that may affect pavement performance was considered initially when identifying the variables. The significant variables that are known to affect performance are pavement layer material properties, subgrade characterisation, climate, traffic, maintenance and drainage. If one or more variables adequately represent each of these parameters, then the model should contain most of the significant variables known to affect performance. A preliminary list of significant explanatory variables that affect pavement performance is prepared from previous literature.

A linear model form of the form y = a0 + a1x1 + a2x2 + … + anxn, typical non-linear power form, that is, y = A0 + A1 X1A2 X2 A3 X3 A4 and y = ex of exponential non-linear, etc., was used. The primary advantage of the general power form and exponential form of the nonlinear models is that they do not rely on fixed exponents, such as squares or cubic powers. An iterative linear and nonlinear regression analysis of various forms of models, viz. linear, power, exponential and logarithmic forms, was used for the development of deterioration prediction. The best model was chosen based on the significance tests, that is, the standard error (SE), t-test, F-test and coefficient of determination of the model R2 value. Basically, SE is one of the many ways to quantify the difference between an estimator and the actual value of the quantity being estimated [5].

3.1.1 Pavement cracking and pothole models

Pavements tend to crack at some point in their life under the combined effect of traffic and environment. There are a number of failure mechanisms associated with the occurrence of cracking in asphalt pavements, such as fatigue, longitudinal cracking, transverse cracking. In this study, cracking represents all types of cracking, such as fatigue cracking, longitudinal cracking, transverse cracking. The model was developed at the project level, and the network level for cracking and potholes is presented below.

$$\begin{array}{ll}{\mathit{\text{Craking}}}_{\mathit{\text{Project Level}}}=& \{\left[0.146\times {\left(\mathit{CBR}\right)}^2-2.286\times \left(\mathit{CBR}\right)\right]+\left[7.928\times \mathit{Ln}\left(N\right)\right]\\ & +\left[0.0006\times {\left(\mathit{LL}\right)}^2+0.208\times \left(\mathit{LL}\right)\right]+\left[0.235\times \left(\mathit{PI}\right)\right]\\ & +\left[-0.131\times {\left(\mathit{PAGE}\right)}^2+1.754\times \left(\mathit{\text{PAGE}}\right)\right]+2.241\}\end{array}$$E1

R² = 0.91, F-test = 78.36, SE = 0.926.

(See Table 1)

Parameter	CBR	N	LL	PI	PAGE
t-test values	−1.14	1.36	5.4	5.3	0.29

Table 1.

Statistical analysis for road level.

Where CBR = California Bearing Ratio (%); N = traffic intensity (msa); LL = liquid limit of the subgrade soil (%); PI = plasticity index of the subgrade soil (%); and PAGE = pavement age (years).

$$\begin{array}{ll}{\mathit{\text{Cracking}}}_{\mathit{\text{Network level}}}=& \{\left[0.014\times {\left(\mathit{CBR}\right)}^2-0.209\times \left(\mathit{CBR}\right)\right]+\left[2.533\times \mathit{Ln}\left(N\right)\right]\\ & +[-0.001\times {\left(\mathit{LL}\right)}^2+0.132\times \left(\mathit{LL}\right)+\left[0.771\times \left(\mathit{PI}\right)\right]\\ & +\left[-0.034\times {\left(\mathit{PAGE}\right)}^2+0.431\times \left(\mathit{\text{PAGE}}\right)\right]+\left[0.067\times \left(\mathit{BC}\right)\right]-9.479\}\end{array}$$E2

R² = 0.69, F-test = 261.255, SE = 2.773.

(See Table 2)

Parameter	CBR	N	LL	PI	N	PAGE	BC
T-test values	1.80	−0.49	4.82	1.25	20.01	0.922	1.03

Table 2.

Statistical analysis for district level.

$$\begin{array}{ll}{\mathit{\text{Pothole}}}_{\mathit{\text{Project level}}}=& \{\left[0.003\times {\left(\mathit{CBR}\right)}^2-0.026\times \left(\mathit{CBR}\right)\right]+1.522\times \left(N\right)+0.096\times \left(\mathit{MMP}\right)\\ & +0.387\times \left(\mathit{NOP}\right)+0.169\times \left(\mathit{\text{PAGE}}\right)+0.090\times \left(\mathit{STV}\right)\\ & -0.067\times \left(C\right)+\left[0.166\times {\left(\mathit{LG}\right)}^2-0.364\times \left(\mathit{LG}\right)\right]+0.866\}\end{array}$$E3

R² = 0.92, F-test = 70.133, SE = 0.11.

(See Table 3)

Parameter	CBR	N	MMP	NOP	PAGE	STV	C	LG
T-test values	−0.866	1.92	1.22	16.49	4.1	0.92	3.09	0.61

Table 3.

Statistical analysis for state level.

CBR = Subgrade California Bearing Ratio (%); N = traffic intensity (msa); MMP = mean monthly precipitation (mm); NOP = No. of potholes; PAGE = pavement age (Yrs); STV = stripping value (%); C = camber (%); LG = longitudinal gradient (%).

3.1.2 Development of performance models using ANNs

Artificial neural networks (ANNs) play a crucial role in pavement deterioration modelling due to their ability to capture complex relationships and patterns within large datasets. The importance of artificial neural networks in pavement deterioration modelling lies in their ability to handle complex, nonlinear relationships, adapt to changing conditions and provide accurate predictions based on data-driven insights. These attributes make ANNs valuable tools for improving our understanding of pavement performance and optimising maintenance strategies.

3.1.2.1 Pavement cracking and pothole models

Bituminous surfaces start developing cracks at some point in their service life under the combined action of traffic loading and the environment. The cracks in the surface are defects of a severe nature, which weaken the pavement structure on account of water penetration and are mainly responsible for further deterioration. Cracks once initiated progress in extent and severity and ultimately lead to potholes. The initiation of cracking is defined as the stage when a crack is observed on the pavement surface.

Cracking is a function of various parameters such as CBR, traffic intensity, liquid limit and plasticity index of the soil; mainly in this analysis, the affected parameters are taken as inputs, and the predicted parameter (cracking) is taken as output. With these input and output parameters, by using the ANN technique, the model was developed for road level, district level and state level with the same data that is used in linear regression analysis.

The neural network worksheet will be displayed with the help of the nntool function in MATLAB. In that window, the input and output parameters will be given, and that should be preceded with a training function in order to train the network with different numbers of iterations based on the quality of predictions. It is to be noted here that for three of the project levels, 90%of the data was trained for the prediction analysis, 5%of the data was validated and 5%of the data was tested. Once the whole data is tested, the overall performance of the data will be given with a target R² value. As the iterations in the tool function increase, the outliers in the predicted data decrease, such that there will be high accuracy in the output prediction. For the whole performance in this analysis, the prediction value will be given by the overall R² value with targeted outputs for cracking and pothole output, which are shown in Figures 3 and 4, respectively.

Figure 3.

Cracking output analysis.

Figure 4.

Pothole output analysis.

3.1.3 Development of performance models using genetic programming

Genetic programming is a type of evolutionary algorithm that uses principles inspired by biological evolution to evolve computer programmes to solve specific problems. In the context of pavement performance prediction, genetic programming can play a significant role for several reasons: while genetic programming can offer advantages, it is essential to note that its success depends on proper application, careful tuning and validation. It should be used in conjunction with domain knowledge and other modelling techniques to ensure accurate and reliable pavement performance predictions. In this study, a detailed analysis was carried out to develop the performance models with the same data but with different techniques. With the help of genetic programming, comprehensive project-level and network-level models have been developed for cracking and pothole distresses. The simplified overall GP expression of cracking analysis for the project level and network level is presented below along with output analysis of the cracking in Figure 5.

Figure 5.

Output analysis of the cracking.

3.1.3.1 Pavement cracking and pothole models

$$\begin{array}{ll}{\mathit{\text{Cracking}}}_{\mathit{\text{Project Level}}}=& 0.02631\mathit{LL}+9.466{\mathit{CBR}}^2\ast {\mathit{PI}}^2+5.957\ast \mathit{CBR}\ast \mathit{PI}\\ & +2.587\ast \mathit{CBR}\ast \mathit{TRF}\ast \mathit{PI}+1.00\end{array}$$E4

R² = 0.99, SE = 0.0012, where

$$\begin{array}{ll}{\mathit{\text{Cracking}}}_{\mathit{\text{Network}}}=& -832.3\ast \mathit{LL}\ast {\mathit{TRF}}^2\ast {\mathit{PI}}^3-0.03076\ast \mathit{LL}\ast \mathit{TRF}\ast \mathit{PI}\\ & +1.258\ast \mathit{CBR}\ast \mathit{TRF}+7.043\ast \mathit{CBR}\ast \mathit{PI}-0.163\ast \mathit{CBR}\\ & +0.1141\ast \mathit{LL}+1.004\end{array}$$E5

R2 = 0.99, SE = 0.002;

$$\begin{array}{ll}{\mathit{\text{Pothole}}}_{\mathit{\text{Project level}}}=& 0.1177\mathit{CDI}-0.07674\mathit{TRF}-0.06887C-0.03837N\\ & +6.702\ast C\ast N+0.0754\ast \mathit{CDI}\ast N+0.6862C^2-0.03837{\mathrm{N}}^2\\ & -0.0305\ast C\ast \mathit{TRF}\ast N-0.0305\ast \mathit{TRF}\ast \mathit{CDI}\ast N+1.006\end{array}$$E6

R2 = 0.99, SE = 0.002

$$\begin{array}{ll}{\mathit{\text{Pothole}}}_{\mathit{\text{Network level}}}=& 0.1122\mathit{CDI}-0.159C+0.9968\ast C\ast \mathit{TRF}+6.946\ast C\ast N\\ & -0.1924\ast \mathit{CDI}\ast N+1.005\end{array}$$E7

R2 = 0.99, SE = 0.002.

The output analysis of network-level data runs with the input and output variables with the Cherkassky function to predict the output with variation between actual and predicted data. The whole analysis will be displayed with different graphical representations with R-square values and a set of errors for each analysis; four graphs will be displayed, which show prediction accuracy, statistical value, a summary of the run and a set of error functions which shows overall performance shown in above Figure 6.

Figure 6.

Output analysis of the pothole.

4. Conclusions and way forward

Based on the results obtained from analysis work carried out in the present study, the following conclusions are drawn.

1. The performance models developed at all project and network levels showed a good prediction of pavement performance. It was also found that ANN and genetic programming give better prediction compared to linear regression at a reasonable degree of accuracy.

2. In linear regression analysis, the overall prediction from each distress is not more accurate due to the outlier percentage being higher at all the levels in the data. The statistical values (P-value; F-value) show a degree of accuracy in simple regression analysis with less quality of fit.

3. The ANN, the overall prediction, is better for projects and networks compared to linear regression. At the stage of testing and validation at the project level for all the distress, it was observed that the testing target prediction was more accurate than that of the validation prediction. At the network in the validation of the data, it was observed that the prediction accuracy was more than that of the testing analysis. However, in terms of overall performance, there is a good quality of fit with more precision.

4. The overall prediction at the project level and network level in genetic programming gives much more accuracy than neural networks. The different techniques used in this study are linear regression, ANN and genetic programming. Each one has its advantages and disadvantages, but when we make a comparison among them, genetic programming gives more prediction when compared to the other techniques, which can solve both linear and nonlinear equations with basic programming coding.

As a way forward, India started systematic rural road development late but progressed notably. Along with PMGSY at the Centre-State Governments on their budget for LVRRs. However, mere asset creation does not lead to sustainable development. Nevertheless, this would be possible only when the created asset is systematically and adequately maintained, saving the created asset.

Conflict of interest

The authors declare no conflict of interest.