Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 12

PAVEMENT PERFORMANCE PREDICTION MODEL

FOR ROUGHNESS USING DEEP LEARNING

Neelansh Mathur
CE16B021

Guide:
Dr. B. Krishna Prapoorna
INTRODUCTION
 Pavement Asset Management System: a systematic approach to assess, monitor, and maintain pavement
assets to service traffic throughout the lifecycle

Prioritization Economic Decision on


Data Condition Performance
and evaluation of Maintenance
collection assessment prediction
Optimization alternatives strategies

 Pavement Performance: the relative ability of pavement to serve traffic over a period of time

 Performance criteria: cracking, rutting, roughness, present serviceability index (PSI),pavement


condition index (PCI) etc.

 Performance prediction models are essential to study the deterioration throughout the life-cycle

 Essential for decision making on maintenance interventions

 Types of performance prediction models :


 Deterministic (e.g., linear regression, non-linear regression)
 Probabilistic (e.g., Markov chains, Weibull process)
 Machine learning approach
10/28/2019 2
LITERATURE REVIEW
• Roughness, a functional parameter measures the deviation from intended longitudinal
profile of a road in International Roughness Index

• There are several intrinsic and extrinsic parameters which affect the roughness of a
pavement surface

• Extrinsic parameters include environment(temperature, precipitation, snowfall), traffic


(ESAL, vehicle speed, load repetitions, truck traffic percentage)

• Intrinsic parameters include construction materials(e.g., concrete, asphalt), pavement


layers, layer thickness, and maintenance strategy

• Over years several models for pavement prediction based on roughness have been
developed
10/28/2019 3
LITERATURE REVIEW
Author Year Model Parameters
Maintenance Expenditure level
Al-Suleiman et al. 1989 Regression Model Pavement age
Traffic loading
Traffic
Environment
Kerali 2000 Non - linear regression model Pavement structure
Pavement surface
Rut depth
{IRI} = (43 + 1.8 × AGE) + {(0.0004 × AADT) × Age
Gulen ey al. 2001
(0.0254/1.60934)} AADT
Mix design factors
Temperature
Prozzi and Madanat 2003 Recurssive non-linear model Moisture
Traffic load
Thickness of different layers
Permanent deformation
Thermal cracking
LTPP 2004 Non-linear regression Model Fatigue cracking
Materials used
Local environment factors
Age
AADT
IRI = 0.34 + 0.059×eAGE/25 × log(AADT) + {0.006 × P200}
Nassiri et al. 2011 Transverse cracking
+0.088 × TransCrk} + 0.013 × MiscCrk} + 0.07 × Rutting
10/28/2019 Miscellaneous cracking 4
Rutting
LITERATURE REVIEW
Author Year Model Categories Range
Wet-cold climate and poor or very poor subgrade
Wet-warm climate and poor or very poor subgrade
Zone
Dry-cold climate and good or very good subgrade
Dry-warm climate and good or very good subgrade
Thick asphaltic concrete
Pavement
Medium thickness asphaltic concrete
family
Thin asphaltic concrete
ln{IRI/IRI(n) - IRI} = b3 + Pavement age
Dalla Rosa et al. 2017
b2 × (e^[AGE×b1]) Low equivalent single axel load—ESAL (average
20 years)
Traffic Medium equivalent single axel load—ESAL (average
loading 20 years)
Heavy equivalent single axel load—ESAL (average
20 years)
Functional Rural road
system Urban road

10/28/2019 5
RESEARCH GAP
• Pavement age and traffic are used in deterministic models
• Previous maintenance history has not been considered
• Markov chains doesn’t consider timeseries data for condition prediction
• Machine Learning models have showed overfitting to trained data and less accuracy on testing
dataset.
• Time complexity of machine learning models are high

OBJECTIVE
• To develop a performance prediction model for roughness using Deep Learning Techniques
• To assess the suitability of the model to Indian conditions
• To further develop performance prediction models for other criteria such as PCI, rutting, cracking
etc.

10/28/2019 6
PRELIMINARY WORK
• Long-Term Pavement Performance (LTPP) database is chosen to
extract data
• Illinois, California and Alabama states were considered for the study
• Section summary reports were summarized to create database
• Preliminary processing was performed to exclude missing data
• For identifying critical parameters, correlation matrices were
constructed
• Factors were categorized based on correlation coefficient (CC) into
• Highly relevant factors (CC > 0.5)
• Fairly relevant factors ( 0.1<CC<0.5)
• Least relevant factors (CC<0.1)
PRELIMINARY WORK
• Highly relevant factors: AADT, AADTT, annual average precipitation,
rutting, transverse cracking, faulting
• Fairly relevant factors: Annual average temperature, freezing index,
deflection
• Least relevant factors: transverse cracking (rigid pavements)

10/28/2019 8
PRELIMINARY WORK
Section A340 Section B340

Annual Annual Annual Annual


Annual
Average Average 18-Kip Average Average 18-Kip
Average
Daily Daily Truck ESAL Daily Daily Truck ESAL
Precipitatio
Traffic Traffic (KESAL) Traffic Traffic (KESAL)
n (mm)
(AADT) (AADTT) (AADT) (AADTT)

0.67813 0.97224 0.998836 0.99916 0.94061 0.93044 0.940814

Section 4082 Section 0601

Annual Annual Annual


Spalling of Avg
Average Average 18-Kip Average Transverse Load Transfer
Faulting Trans. Annual Deflection
Daily Daily Truck ESAL Daily Truck Cracking Spalling of Efficiency of Transverse
(mm) Joints 18-Kip Average Transverse (9-Kip,
Traffic Traffic (KESAL) Traffic (Count) Rutting Trans. Joints (%)
(Count) ESAL Daily Truck Cracking wheel load
(AADT) (AADTT) (AADTT) (mm) Joints
(KESAL) Traffic (Count) ) at 0" from
(Count)
(AADTT) Load Plate
(microns)

Approach Leave
0.964419 0.969425 0.96855 1 1 -1
-0.54421 0.68626 0.7263 -0.61386 0.56219 -0.6818 -1 -1
10/28/2019 9
PRELIMINARY WORK
Section 0601 Section 4082
Climate (Virtual Weather Station (VWS) Data)

Transverse
Corner Breaks Annual Average Freeze Index
Annual Average Annual Average Annual Average Freeze Cracking
(Count) (deg C deg days)
Precipitation (mm) Temperature (deg C) Index (deg C deg days) (Count)

-0.37154 0.039445 -0.1664 -0.2871 -0.30054 0.31559

Section B340

Annual Average Annual Average Annual Average Freeze


Precipitation (mm) Temperature (deg C) Index (deg C deg days)

-0.145454 0.11148 0.152488

10/28/2019 10
FUTURE PLAN
Pavement
Performance
Prediction Model

Testing model for Testing in Indian


Data collection Database creation Develop DL model
accuracy conditions

Test the developed Using data from


Summarize the excel
LTPP Database model with the NHAI to test the
data for every section
testing dataset model

Identify critical Test with datasets


Sections in India
factors from other sources

Split the dataset for


training and testing
REFERENCES
• Chen, D., and Mastin, N. (2016). "Sigmoidal Models for Predicting Pavement Performance
Conditions". Journal of Performance of Constructed Facilities, 30(4), 04015078.
• Park, S., and Kim, J. (2019). "Comparative Analysis of Performance Prediction Models for
Flexible Pavements". Journal of Transportation Engineering, Part B: Pavements, 145(1),
04018062.
• George K.P., Rajagopal, A.S., and Lim L.K. (1989). “Models for Predicting Pavement
Deterioration”. Transportation Research Record, 1215.
• Gupta, A., Kumar, P., and Rastogi R. (2014). “Critical Review of Flexible Pavement Performance
Models”. KSCE Journal of Civil Engineering, 18(1), 142-148.
• Mahmood M., Rahman, M., and Mathavan, S. (2016). “Distress Based Pavement Performance
Prediction Models”. Proceedings of the Eighth International Conference on Maintenance and
Rehabilitation of Pavements, 358-368.

10/28/2019 12

You might also like