Although vehicles emissions have a very significant impact on CO2 emissions, there remains a level of uncertainty concerning the methodological assumptions and parameters to consider in the calculation of greenhouse gas (GHG) emissions coming from the use phase of road pavements (Trupia et al 2016). In fact, recent studies highlighted how existing models can lead to very different results and that because of this, they are not fully ready to be implemented as standard in the life-cycle assessment (LCA) framework (Santero et al 2011; Trupia et al 2016). This study presents an innovative approach, based on the application of Machine Learning to ‘Big Data’, for the calculation of the use phase emissions of road pavements due to truck fleet fuel consumption. The study shows that the Machine Learning regression technique is suitable to analyse the large quantities of data, coming from fleet and road asset management databases effectively, assessing and estimating the impact of rolling resistance-related parameters (pavement roughness and macrotexture measurements) on the use phase in road pavement LCA.

1. Reducing Uncertainty in Structural Safety
Special Session SS6
Ghent, Belgium
28-31 October 2018

2. A machine learning
approach for the estimation
of fuel consumption related to
road pavement rolling res.
for large fleets of trucks
Federico Perrotta, Tony Parry, Luis C. Neves,
Mohammad Mesgarpour
NTEC
Nottingham Transportation
Engineering Centre
HORIZON 2020
29/10/2018 – IALCCE 2018, Ghent, Belgium

3. Outline HORIZON 2020
 Introduction
 What does impact fuel consumption?
 In terms of money
 A new approach
 Previous studies
 Big Data
 Modelling
 A UK case study
 Which variables to consider?
 Artificial Neural Network (ANN)
 Parametric analysis
 Conclusions

4. Introduction HORIZON 2020
In UK, the road infrastructure is
the most extensive and valuable
asset (House of Commons, 2011):
 ~ 295,000 km;
 ~ £344 billion;
 ~ £4 billion per year for
maintenance and repair.
“Ensuring a good state of the road infrastructure is critical for our
economy and society.” - European Parliament, 2014
However, due to the use of oil derivative fuels, currently, the road
transport sector is also a major contributor to the production of
greenhouse gases.

5. What does fuel consumption depends on? HORIZON 2020

6. In terms of money HORIZON 2020
With accurate fuel consumption estimates and a review of the
current road maintenance strategies we can actually make a
difference!
In the USA every year: 200
billion gal of motor fuel are
consumed
for e.g. 2% that would be
between $5 and $10 billion
to re-invest in maintenance*
Previous studies (e.g. Chatti and Zaabar 2012) said that road
roughness can impact fuel consumption by up to 5%.
* Considering the current cost of Diesel

7. Previous studies HORIZON 2020
What happens instead at the network level under real driving
conditions?
 Experimental approach;
 Limited number of vehicles;
 Selected road segments;
 Under controlled conditions (e.g. const. speed).
?

8. Big Data HORIZON 2020
(SAE J1939 – data from sensors installed
on modern trucks inform fleet managers)
(HAPMS – (Highways Agency Pavement
Management System) database containing
information for road managers)

9. A UK case study
~ 300 km of motorway
Considering a week in October ‘16
473 medium trucks
5,423 records
asphalt and concrete
M1
M1
M18
Euro 6 with 4 axles
1 minute or 1 mile
Fuel used in 0.001 l

10. Which variables to consider? HORIZON 2020
From 56variables
initially available

11. HORIZON 2020
14variables selected
Which variables to consider?
BA (Boruta Algorithm): based on
the theory of decision trees and
random forests:

12. Artificial Neural Network (ANN) HORIZON 2020
Grid-search method and cross-validation can be used to
define the optimal structure of the network.
Using the resilient backpropagation algorithm with
backtracking (rprop+):

13. Artificial Neural Network (ANN)
Avg. of 10-fold cv
R2 0.88
RMSE 4.02 l/100km
MAE 2.59 l/100km
Measured FC Estimated FC Error
24.73 l/100km 24.35 l/100km -1.5%
Using:
3,904 records for training
1,301 for validation
218 to test the model
For the test set:
On average:

14. Parametric analysis HORIZON 2020
Perrotta F., Parry T., Neves L.C., Mesgarpour M., Benbow E. and Viner H., A big
data approach for investigating the performance of road infrastructure. Civil
Engineering Research in Ireland (CERI) 2018. Dublin, Ireland.

15. Conclusions HORIZON 2020
 BA demonstrated to be able to identify significant variables out of the
large datasets quickly and effectively;
 NN demonstrated to be able to estimate the fuel consumption of the
considered fleet of trucks accurately;
 Other studies (Perrotta et al. 2018) showed how a parametric analysis can
be used to estimate impacts of each of the variables included in the
developed model;
 Once fuel consumption is known, it is possible to estimate eqCO2 or
quantity of GHGs produced using the emissions factors published by EPA
(2018) or other environmental agencies;
 Overall it is possible to say that the ‘Big Data’ approach seems to be
promising to estimate costs and environmental impact from the use-
phase of the road infrastructure;
 Further research should focus on more accurate data, a wider range of
vehicles, different road materials, weather conditions, urban areas, etc.
which will improve applicability of the study and reliability of the
developed models.

16. What is next? HORIZON 2020
The answer is in data!

17. Questions?
Thank you for your
attention!
esr13truss.
blogspot.co.uk
This project has received funding from the European Union’s Horizon
2020 research and innovation programme under the Marie
Sklodowska-Curie grant agreement No. 642453
NTEC
Nottingham Transportation
Engineering Centre
HORIZON 2020
trussitn.eu
Tony.Parry@nottingham.ac.uk
Federico.Perrotta@nottingham.ac.uk