Federico Perrotta

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.

Keywords: Fuel Consumption, GHG emissions, Pavement LCA, Big Data, Machine Learning

“Using truck sensors for road pavement performance investigation” is a research project within TRUSS, an innovative training network funded from the EU under the Horizon 2020 programme. The project aims at assessing the impact of the condition of the road pavement unevenness and macrotexture, on the fuel consumption of trucks to reduce uncertainty in the framework of life-cycle assessment of road pavements. In the past, several studies claimed that a road pavement in poor condition can affect the fuel consumption of road vehicles. However, these conclusions are based just on tests performed on a selection of road segments using a few vehicles and this may not be representative of real conditions. That leaves uncertainty in the topic and it does not allow road mangers to review the current road maintenance strategies that could otherwise help in reducing costs and greenhouse gas emissions from the road transport industry. The project investigated an alternative approach that considers large quantities of data from standard sensors installed on trucks combined with information in the database of road agencies that includes measurements of the conditions of the road network. In particular, using advanced regression techniques, a fuel consumption model that can take into consideration these effects has been developed. The paper presents a summary of the findings of the project, it highlights implications for road asset management and the road maintenance strategies and discusses advantages and limitations of the approach used, pointing out possible improvements and future work.

In the past, experimental studies have estimated the impact of road surface conditions on vehicle fuel consumption to be up to 5%. This, together with a review of the current maintenance strategies, could lead to a significant reduction of costs and GHG emissions from the road transport industry. However, this has been established in experiments using a limited number of instrumented test vehicles under carefully controlled conditions (e.g. steady speed, no gradient etc.) and for short test sections. What is less clear is the significance of these impacts on vehicle fleet fuel economy, under real driving conditions and at network level. This has recently gained more focus in the highway authority and research community as carbon footprinting of road maintenance plans has gained importance. Modern trucks are fitted with many sensors as standard and used to inform decisions on maintenance and driver training requirements in large fleets. However, much of the information produced could also be used in the measurement of how road condition influences performance in terms of vehicle operation. In particular, one research questions has not been sufficiently well answered: ‘what is the influence of road pavement roughness on truck fleet fuel consumption?’ The research aims to provide an answer to this questions to help prioritize pavement maintenance and design decisions with respect to user and environmental impacts. In particular, the research will develop a fuel consumption model that would help engineers in assessing the impact of road conditions on truck fleet fuel consumption. Some of the most innovative regression techniques, including random forests and artificial neural networks will be used for the purpose. It is expected that the developed tools will help in reducing uncertainties in the topic and extend the system boundaries of life-cycle carbon footprint of the current road maintenance strategies.

There is a multitude of models available to assess structural safety based on a set of input parameters. As the degree of complexity of the models increases, the uncertainty of their output tends to decrease. However, more complex models typically require more input parameters, which may contain a higher degree of uncertainty. Therefore, it becomes necessary to find the balance that, for a particular scenario, will reduce the overall uncertainty (model + parameters) in structural safety. The latter is the objective of the Marie Skłodowska-Curie Innovative Training Network titled TRUSS (Training in Reducing Uncertainty in Structural Safety) funded by the EU Horizon 2020 research and innovation programme (http://trussitn.eu). This paper describes how TRUSS addresses uncertainty in: (a) structural reliability of materials such as basalt fiber reinforced polymer, (b) testing techniques in the assessment of concrete strength in buildings, (c) numerical methods in computing the non-linear response of submerged nuclear components subjected to an earthquake, (d) estimation of life of wind turbines, (e) the optimal inspection times and management strategies for ships, (f) characterization of the dynamic response of ship unloaders and (g) the relationship between vehicles fuel consumption and pavement condition.-> Link to full text in repository

 European infrastructure is extensive and well developed, but much of it is also aging. Replacing this aging network is costly, disruptive and time-consuming – so it is vital that what is already in place is utilized to its full potential. This scenario is made even more perilous by the reduced spending in maintenance caused by the economic downturn of recent years. As such, a management strategy that guarantees maintenance and structural safety with the best use of the resources available is required. This goal is at the heart of the TRUSS Innovative Training Network (ITN). TRUSS addresses a lack of expertise on the management and modernization of an aging infrastructure stock that is critical for society to function and prosper. The cost of making repairs once an infrastructure starts to fail is prohibitive, yet there is no easy way to measure how infrastructure deteriorates over time and to assess structural safety.  -> External link to full paper as published online. [DOI] -> External link to full text as published online -> Link to full text in repository 

The paper presents a comparison of fuel consumption estimates computed using the HDM-4 model with measurements from a large database owned by truck fleet managers. The study considers data from three different types of truck and 1,645 vehicles in total for a case study in the United Kingdom (UK). Typical fuel consumption has been estimated using HDM-4 for light, medium and heavy trucks. Trucks were considered driving at 85km/h on a completely flat 1 km road segment. This has been compared to the average fuel consumption of the same types of trucks driving on the M18 motorway in England. The model has been configured based on a calibration of HDM-4 for the local conditions of the UK in 2011 (12). The paper shows that application of the model in the current state of calibration does not always predict correctly what actually happens in real conditions at local level. From the analysis of the information available from the fleet manager database, the paper tries then to address the differences between measurements and predictions of the model by making assumptions about the factors that impact the final results, aiming to develop a solution for more precise and reliable estimates. The authors suggest that, in order to use the model at a strategic level, a recalibration of the HDM-4 model and further analysis should be performed in the UK. Finally, the paper discusses the methodology used for developing HDM-4 (and similar models) and suggests possible alternatives for future work regarding the calibration of HDM-4 and modeling of the fuel consumption of road vehicles using ‘Big Data’.

This paper presents the application of three Machine Learning techniques to fuel consumption modelling of articulated trucks for a large dataset. In particular, Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN) models have been developed for the purpose and their performance compared. Fleet managers use telematic data to monitor the performance of their fleets and take decisions regarding maintenance of the vehicles and training of their drivers. The data, which include fuel consumption, are collected by standard sensors (SAE J1939) for modern vehicles. Data regarding the characteristics of the road come from the Highways Agency Pavement Management System (HAPMS) of Highways England, the manager of the strategic road network in the UK. Together, these data can be used to develop a new fuel consumption model, which may help fleet managers in reviewing the existing vehicle routing decisions, based on road geometry. The model would also be useful for road managers to better understand the fuel consumption of road vehicles and the influence of road geometry. Ten-fold cross-validation has been performed to train the SVM, RF, and ANN models. Results of the study shows the feasibility of using telematic data together with the information in HAPMS for the purpose of modelling fuel consumption. The study also shows that although all the three methods make it possible to develop models with good precision, the RF slightly outperforms SVM and ANN giving higher R-squared, and lower error. -> Link to full text in repository

Considering data from 260 articulated trucks, with ∼12900 cc Euro 6 engines driving along a motorway in England (M18), the study first shows how different approaches lead to the conclusion that road pavement surface conditions influence fuel consumption of the considered truck fleet. Then, a multiple linear regression for the prediction of fuel consumption was generated. The model shows that evenness and macrotexture can impact the truck fuel consumption by up to 3% and 5%, respectively. It is a significant impact which confirms that, although the available funding for pavement maintenance is limited, the importance of limiting GHG emissions, together with the economic benefits of reducing fuel consumption are reasons to improve road condition (Zaabar & Chatti, 2010). [DOI] -> Link to full text in repository

In Europe, the road network is the most extensive and valuable infrastructure asset. In England, for example, its value has been estimated at around £344 billion and every year the government spends approximately £4 billion on highway maintenance (House of Commons, 2011).

Fuel efficiency depends on a wide range of factors, including vehicle characteristics, road geometry, driving pattern and pavement condition. The latter has been addressed, in the past, by many studies showing that a smoother pavement improves vehicle fuel efficiency. A recent study estimated that road roughness affects around 5% of fuel consumption (Zaabar & Chatti, 2010). However, previous studies were based on experiments using few instrumented vehicles, tested under controlled conditions (e.g. steady speed, no gradient etc.) on selected test sections. For this reason, the impact of pavement condition on vehicle fleet fuel economy, under real driving conditions, at network level still remains to be verified.

A 2% improvement in fuel efficiency would mean that up to about 720 million liters of fuel (~£1 billion) could be saved every year in the UK. It means that maintaining roads in better condition could lead to cost savings and reduction of greenhouse gas emissions.

Modern trucks use many sensors, installed as standard, to measure data on a wide range of parameters including fuel consumption. This data is mostly used to inform fleet managers about maintenance and driver training requirements. In the present work, a ‘Big Data’ approach is used to estimate the impact of road surface conditions on truck fleet fuel economy for many trucks along a motorway in England. Assessing the impact of pavement conditions on fuel consumption at truck fleet and road network level would be useful for road authorities, helping them prioritize maintenance and design decisions. -> Link to full text in repository

Experimental studies have estimated the impact of road surface conditions on vehicle fuel consumption to be up to 5% (Beuving et al., 2004). Similar results have been published by Zaabar and Chatti (2010). However, this was established testing a limited number of vehicles under carefully controlled conditions including, for example, steady speed or coast down and no gradient, amongst others. This paper describes a new “Big Data” approach to validate these estimates at truck fleet and route level, for a motorway in the UK. Modern trucks are fitted with many sensors, used to inform truck fleet managers about vehicle operation including fuel consumption. The same measurements together with data regarding pavement conditions can be used to assess the impact of road surface conditions on fuel economy. They are field data collected for thousands of trucks every day, year on year, across the entire network in the UK. This paper describes the data analysis developed and the initial results on the impact of road surface condition on fuel consumption for journeys of 157 trucks over 42.6km of motorway, over a time period of one year. Validation of the relationship between road pavement surface condition and vehicle fuel consumption will increase confidence in results of LCA analyses including the use phase. [DOI] -> Link to full text in repository