Artificial Neural Networks in Pavement Engineering: A Recent Review

Abhishek Kurian*, Elvin Sunildutt

Department of Civil Engineering, Saintgits College of Engineering, Kottayam, Kerala
*Corresponding author: abhishekkurian99@gmail.com
doi: https://doi.org/10.21467/proceedings.112.66

ABSTRACT
The application of Artificial Neural Networks (ANN) in civil engineering has increased drastically in the
past few years. ANN tools are nowadays used commonly in developed countries over various fields of civil
engineering like geotechnical, structural, traffic, pavement engineering etc. This paper deals with the review
of recent advancements and utilization of ANNs in pavement engineering. The review will focus on
pavement performance prediction, maintenance strategies, distress intensity detection through deep
learning techniques, pavement condition index prediction etc. The use of ANNs in pavement management
systems are expected to furnish a systematic schedule and economic management strategies in the field of
pavement engineering. The use of ANNs combined with deep learning techniques help to address complex
problems in pavement engineering and pave the way to a sustainable future.

Keywords: Artificial Neural Networks, deep learning

1 INTRODUCTION
Recent studies have shown that Artificial Neural Networks play a significant role in Pavement Management
System (PMS). PMS is a pressing priority especially in developing countries like India for efficient and economic
construction and maintenance of pavements. PMS was introduced as a tool to ensure proper maintenance and
rehabilitation strategies for pavement construction projects with a low budget (Kirbas 2010). PMS helps to
manage different stages of roads pavement such as planning, designing, construction and maintenance (S. Choi
et. al. 2019). Proper execution of PMS can ensure efficient budget distribution, project prioritization and
economical maintenance strategies which can maximize the service life of a pavement (Fawaz 2018). The
development of technologies like “machine learning” and “deep learning” has further improved the scope of
artificial intelligence in the field of pavement engineering. Deep learning is a technique which builds complex
artificial neural networks based on extensive data (Hern 2016). In simple words deep learning and machine
learning resemble the ability of humans to perform tasks based on knowledge and experience. These techniques
are being used recently for crack detection, Pavement Condition Index (PCI) prediction and effective
maintenance. According to recent studies nonlinear calculations in pavement engineering are being solved using
ANN models (Ceylan et. al. 2014). Optimization by feeding adequate data to an ANN model can improve
accuracy. This paper deals with applications of deep learning ANN models in pavement engineering ranging
from performance prediction, pavement condition prediction, distress detection using image sensing techniques
and management strategies.

1.1 ARTIFICIAL NEURAL NETWORKS

ANN work is based on the basic principle of human nervous system and consists of neurons which are
interconnected. Neurons are distributed in three layers such as input layer, hidden layer and output layer. ANN
models have several features such as adaptivity, non-linearity, mapping and can deal with noisy data (M.

© 2021 Copyright held by the author(s). Published by AIJR Publisher in “Proceedings of International Web Conference in Civil Engineering for a
Sustainable Planet” (ICCESP 2021) March 5th–6th, 2021. Organized by Habilete Learning Solutions, Kollam, Kerala, India in collaboration with
American Society of Civil Engineers (ASCE); TKM College of Engineering; and Baselios Mathews II College of Engineering, Kollam, Kerala, India.
Proceedings DOI: 10.21467/proceedings.112; Series: AIJR Proceedings; ISSN: 2582-3922; ISBN: 978-81-947843-3-3
Artificial Neural Networks in Pavement Engineering: A Recent Review

Abambres et. al. 2017). ANN models require training and validation in order to perform calculations accurately.
A unique data set is used for training the model whereas another data set which has no resemblance to the
initial one is used for validation purposes. However, when the data is insufficient, the accuracy level of the
model suffers evidently and parameter interpretation becomes difficult. During field data measurement, factors
such as climate and quality of construction are difficult to be taken into account. Hence there is a need for
nonlinear models to be established in order to get accurate results.

Figure. 1. ANN model (Thube 2012)

1.2 OBJECTIVE
ANNs act as an alternative technique when considering the conventional techniques. The conventional
techniques are limited or restricted by linearity, normality variable independence etc. Such limitations in
conventional methods along with the rising potential of artificial intelligence and deep learning had encouraged
various researchers to develop ANN models for prediction, evaluation and assessment of pavement
performance and condition. The objective of this study is to provide a detailed review of the recent
advancements and development of ANNs in the field of pavement engineering.

2 APPLICATIONS
2.1 PREDICTION OF PAVEMENT CONDITION INDEX (PCI)
Pavement Condition Index (PCI) is a rating system ranging from 0 to 100 (0-failed, 100-good) which portrays
the distress intensity of a pavement in terms of a numerical index (ASTMD 6433:18). Pavement distress
assessment is necessary to establish a robust PMS to ensure cost effective practices and efficient maintenance
strategies. Usually, such distress intensity assessment of pavements is carried out by a panel of experts working
in the field of pavement engineering. Distress detection is performed using Network Survey Vehicles (NSV) or
on foot surveys. Such human dependent assessment is time consuming and uneconomical. H. Majidifard et. al.
2020 developed a deep learning model which uses image sensing to detect the type and intensity of cracks.
More than 7000 google street view images were used to train the model for crack detection. The types of distress
Proceedings DOI: 10.21467/proceedings.112 Series: AIJR Proceedings
554
ISBN: 978-81-947843-3-3 ISSN: 2582-3922
 Kurian, A., & Sunildutt, E., AIJR Proceedings, pp.553-558, 2021

were divided into nine classes such as reflective crack, transverse crack, block crack, longitudinal crack, alligator
crack, sealed reflective crack, lane longitudinal crack, sealed longitudinal crack and potholes. A hybrid model
was developed using already existing models like YOLO and U-net. Further training was done using google
street view images to make the model more human independent and accurate. The developed model was found
to be effective, economical and time saving. Further study needs to be conducted in order to develop this model
into a software to perform all steps like image extraction, analysis and PCI prediction as a whole.

Figure. 2. Distress detection using deep learning model (H. Majidifard et. al. 2020)

In developing countries like India, there is huge need for PCI prediction for rehabilitation and maintenance
works. However, google street view function is not fully established in India due to certain restrictions put forth
by the government. These restrictions are expected to be removed in the future, meanwhile manually taken
images using drones or other equipment can be used to get results from such models. A. Jiminez et. al 2016
developed ANN models to predict PCI values based on data such as Equivalent Standard Axle Load (ESALs),
Structural Number (SN), average annual daily traffic, age of the pavement and thickness. PCI values for both
flexible and rigid pavement were calculated using this model. H. Shahnazari et. al 2012 compared the results of
ANN models and genetic programming method. PCI calculation program was developed using both these
techniques and was found to be a reliable prediction tool in PMS. ANN models were found to give more
accurate values of PCI. M. Jalal et. al. 2017 developed an optimal ANN model to predict PCI values and input
variables such as pavement type and year. The optimal model was found to have a lower value of error
compared to the normal ANN models. The coefficient of determination of the entire model was found to be
high.

2.2 PREDICTION OF PAVEMENT PERFORMANCE

Prediction of pavement performance adds to the effectiveness of a PMS. Predicting pavement performance
helps in proper budget allocation especially in scheduling maintenance strategies and rehabilitation works
throughout the service life of a pavement. Pavement performance prediction model in PMS can be defined as
“a mathematical description of the expected values that a pavement attribute will take during a specified analysis
period” (Hudson et. al., 1979). Such models help to schedule rehabilitation works and maintenance in time
(George 2000).

Proceedings of International Web Conference in Civil Engineering for a Sustainable Planet (ICCESP 2021)

555
Artificial Neural Networks in Pavement Engineering: A Recent Review

Figure. 3. Pavement performance (Haas et. al. 1994)

Pavement performance models in PMS includes deterministic, probabilistic, expert and ANN models (Wolters
et. al. 2010). ANN models can be used to develop meaningful correlation between input variables (Rafiq et. al.
2001). Gencel et. al 2011 developed ANN models and linear regression models to correlate cement content,
metal content and traffic loading on rigid pavements. ANN model was found to be more accurate while
predicting wear of concrete. F. Alharbi 2018 developed a Multiple Linear Regression (MLR) model and ANN
model to predict pavement performance. Relative contribution of input variable on distress or condition index
was analyzed. A reliable correlation between structural capacity and rutting was developed using an ANN model
and was found to have high R2 values which further validates the result obtained. Influence of climate data in
addition to historical data was included in this study. Weather factors such as average annual temperature,
snowfall and rainfall was found to affect the performance of pavements. The study determined the influence
of weather on pavement deterioration.
Ceylan et al. 2005 developed ANN models to predict strains using FWD deflection data. Tapkin et al. 2012
developed ANN models to predict stiffness, strain buildup, stability and marshall quotient. Stress, strain and
deflection in rigid pavements were predicted using ANN models (Saleh 2015). Longitudinal strains in flexible
pavements subjected to moving loads were predicted using ANN models by Shafabakhsh et al. 2015. I.N.
Abdallah et al. 2009 developed ANN models to predict critical strains using deflection bowl obtained from
Falling Weight Deflectometer. The input variables such as thickness of pavement and pavement deflection data
was used in the ANN model for accurate performance prediction. Comparison of life of pavement calculated
using ANN model and conventional methods like back calculation method were performed in this study.
Pavement properties such as layer moduli, thickness, poison’s ratio, stress, strain, deflection etc. are taken into
consideration while designing a pavement. An efficient pavement design aids the service life of a pavement.
Ceylan et al. (2007) predicted moduli values for each layer using input variables such as deflection data from
FWD and thickness of the layer. Bayrak et al. 2008 used ANN models to backcalculate pavement properties
using FWD data. Pożarycki (2015) predicted layer thickness of bituminous pavements using input variables
such as deflection data, stress applied and temperature.

Proceedings DOI: 10.21467/proceedings.112 Series: AIJR Proceedings

556
ISBN: 978-81-947843-3-3 ISSN: 2582-3922
 Kurian, A., & Sunildutt, E., AIJR Proceedings, pp.553-558, 2021

2.3 PREDICTION OF PAVEMENT DISTRESS

The intensity of distress on pavements influence ride comfort, fuel consumption, frequency of accidents, fatigue
etc. Solhmirzaei et al. 2012 developed an ANN model to predict pavement distress. Input variables such as
vehicular acceleration and output variable such as vertical displacement was used to train the model. H.
Majidifard et. al. 2020 used image sensing techniques for detecting pavement distress. Further studies need to
be conducted for sensing distress such as rutting. Ceylan et al. 2011 developed ANN models to generate
accurate stress intensity factor (SIF) which occurs due to changes in temperature. Thube 2012 developed ANN
models to predict accumulation of pavement distress. Data including intensity of rutting, raveling and cracking
area was used as model outputs. Yoo et al. 2016 developed models to differentiate noise objects and pavement
distress in images. H. Majidifard et al. 2020 used a similar technique to get effective results. Saghafi et al. 2009
predicted faulting in rigid pavements by taking factors such as age of a pavement and condition of layers into
account. Wu et al. 2014 detected intensity of fatigue and reflective cracking using ANN models.

3 CONCLUSION
Development of ANNs have a significant impact in the performance as well as rehabilitation aspects of
pavement engineering. With its evolution, classification of pavement distress on the basis of its type and
intensity are undertaken relatively with ease. Pavement parameters such as PCI can be quantified and pavement
rating can be developed with minimal effort for rehabilitation purposes. Computation of pavement
performance is easily one of the highlights of ANN models. ANN models can be an efficient analysis and
predicting tool once they are trained and developed with appropriate datasets. The predictions made using
ANNs can be utilized for enhancements subjected to improving pavement performance. Despite the
advancements and applications of artificial intelligence in transportation engineering, there is a huge gap that
needs to be filled in the branch of pavement engineering. Many researchers recently have introduced the
concept of machine learning and deep learning into their studies related to pavement engineering. Such tools
can be used to design and develop robust pavements with maximum service life. ANN models may take hours
or certain months to train along with collection of immense data. However, time is a trade-off when it comes
to scope and opportunity it puts forward to this field.

How to Cite this Article:

Kurian, A., & Sunildutt, E. (2021). Artificial Neural Networks in Pavement Engineering: A Recent Review.
AIJR Proceedings, 553-558.

REFERENCES
AASHTO. (1993). AASHTO guide for design of pavement structures. 1993 (Vol. 1). AASHTO.
ASTM D6433-20, Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys, ASTM International, West Conshohocken,
PA, 2020
Amin, S. and Amador-Jiménez, L., (2016). Backpropagation Neural Network to estimate pavement performance: dealing with measurement errors,
Road Materials and Pavement Design, 1-21, DOI: 10.1080/14680629.2016.1202129
Ceylan, H., Bayrak, M. and Gopalakrishnan, K. (2014). Neural Networks Applications in Pavement Engineering: A Recent Survey, International
Journal of Pavement Research and Technology, 7(6), 434-444.
Ceylan, H., Gopalakrishnan, K., and Kim, S. (2014). I-BACK: Iowa‘s Intelligent Pavement Backcalculation Software. Geo-Congress 2014
Technical Papers, Geo-Technical Special Publication (GSP), No. 234, pp. 2952-2959.
Choi, S. and Do, M. (2019) Prediction of Asphalt Pavement Service Life using a Deep Learning Method. International Journal of Highway
Engineering, 20, 57–65.
Eldin, N.N. and Senouci, A.B. (1995). Use of neural networks for condition rating of jointed concrete pavements. Advances in Engineering
Software, 23(3), pp. 133-141.

Proceedings of International Web Conference in Civil Engineering for a Sustainable Planet (ICCESP 2021)

557
Artificial Neural Networks in Pavement Engineering: A Recent Review

Gopalakrishnan, Ceylan, H., and Attoh-Okine. (2009) Intelligent and Soft Computing in Infrastructure Systems Engineering: Recent Advances.
Studies in Computational Intelligence. Springer, pp. 47–66, Heidelberg, Berlin, Germany.
Gajewski, J. and Sadowski, T. (2014). Sensitivity analysis of crack propagation in pavement bituminous layered structures using a hybrid system
integrating Artificial Neural Networks and Finite Element Method. Computational Materials Science, 82, pp. 114-117.
Haas, R., W. R. Hudson, & J. Zaniewaski. (1994). Modern pavement management. Malabar, FL: Krieger.
Kirbas, U. and Karasahin, M. (2016) Performance models for hot mix asphalt pavements in urban roads. Construction and Building Materials 116,
281–288.
Kirbas, U. and M. Gursoy. (2010). Developing the basics of pavement management system in Besiktas district and evaluation of the selected
sections. Scientific Research and Essays, 5(8), 806-812.
Kim, S., Gopalakrishnan, K., and Ceylan, H. (2009). Neural Networks Application in Pavement Infrastructure Materials. Studies in Computational
Intelligence. Springer, pp. 47–66, Heidelberg, Berlin, Germany.
Majidifard, H., Adu-Gyamfi, Y. and Buttlar, W (2020). Deep Machine Learning Approach to Develop a New Asphalt Pavement Condition Index,
Construction and Building Materials 247 (2020) 118513
Shahnazari, H., Tutunchian, M. A., Mashayeki, M. and Amini, A. A, Application of soft computing for prediction of pavement condition index
(2012). Journal of Transportation Engineering 138(12):1495-1506
Sharma, S. and Das A (2008). Backcalculation of pavement layer moduli from falling weight deflectometer data using an artificial neural network,
Canadian Journal of Civil Engineering, 35(1), 57–66.
Saghafi, B., Hassani A, Noori, R. and Bustos MG (2009). Artificial neural networks and regression analysis for predicting faulting in jointed
concrete pavements considering base condition, International Journal of Pavement Research and Technology, 2(1), 20-25.
Saleh, M., (2015). Prediction of Rigid Pavement Responses under Axle Loads Using Artificial Neural Network, International Journal of Pavement
Research and Technology, 8(1), 10-16.
Santos, J., Ferreira, A. and Flintsch G (2017). An adaptive hybrid genetic algorithm for pavement management, International journal of pavement
Engineering
Tapkın, Çevik, A., and Özcan Ş (2012). Utilizing Neural Networks and Closed Form Solutions to Determine Static Creep Behaviour and Optimal
Polypropylene Amount in Bituminous Mixtures, Materials Research, 15(6), 865-883.
Thube, D. T., (2012). Artificial Neural Network (ANN) Based Pavement Deterioration Models for Low Volume Roads in India. International Journal
of Pavement Research and Technology, 5(2), pp. 115-120.

Proceedings DOI: 10.21467/proceedings.112 Series: AIJR Proceedings

558
ISBN: 978-81-947843-3-3 ISSN: 2582-3922