Exploring the impact of road surface conditions on truck fleet fuel consumption through Big Data

The thesis presents a novel approach for estimating the impact of road roughness and macro-texture on truck fleet fuel consumption based on Big Data. An extensive literature review was carried out to provide a comprehensive background for the study. Objectives of the study are defined as, primarily, to reduce the uncertainties left by previous studies, which based on experimental data (likely not be representative of real driving conditions) claimed that road roughness and macro-texture could affect the fuel consumption of road vehicles. In fact, this may represent an opportunity for road agencies that, in the light of these results, may wish to review the current maintenance strategies of road pavements to save significant direct and indirect costs for the society. Therefore, instead of carrying out new experiments, by exploring opportunities in the large amount of real-world data available in England, from truck fleet managers, road agencies and weather institutions, the method introduced in the present study promises to be able to provide estimates representative of real driving conditions. The method also differs from those presented in the past for its characteristic repeatability and adaptability to new situations. That includes adaptation of the model to different vehicles and to countries other than England, all without performing any additional field tests. This makes the introduced methodology unique. Advanced statistics and machine learning (ML) techniques have been used to analyse the data. In particular, a multiple linear regression (LR), a support vector regression (SVR), a random forest (RF) and an artificial neural network (ANN) model, including the effect of road roughness and macro-texture as longitudinal profile variance (LPV) and sensor-measured texture depth (SMTD) respectively, have been developed for light, medium and heavy trucks driving along motorways and A roads in England. A parametric analysis was then used to interpret the obtained results and quantify the impact that each of the considered variables, including LPV and SMTD, have on the fuel consumption of the considered truck types. Results show that, although the present study confirms that road surface characteristics, such as roughness and macrotexture, can affect the fuel consumption of trucks, due to the low quality of the data available, that is currently difficult to quantify. A comparison of the results obtained with the findings of studies conducted in the past, shows that there is some match in the order of magnitude of the estimates made, but this is not always the case. For instance, the impact of road roughness, measured as LPV at 3, 10 and 30 metres wavelength, was estimated to be between -2.91% and +6.27% for light trucks, between -8.56% and +3.97% for medium trucks and between -6.55% and +6.28% for heavy trucks, while the effect of macrotexture, measured as SMTD, was estimated to be between +1.04% and +1.21% for light trucks, between -1.21% and +5.98% for medium trucks and between -2.33% and +7.84% for heavy trucks. Variance among the obtained estimates is high and therefore, although results of the present study seem to be promising, it is fair to say that, at the current state of technology, while this approach is feasible, it is difficult to estimate the effect of road surface conditions on truck fleet fuel consumption using the Big Data approach, and further investigation is required to optimise the methodology. Future work will have to consider a wider range of road conditions, increased vehicle reporting frequencies, vehicle speed and vehicle types, including cars. This will extend the applicability of the study and is necessary before estimation of the impact of road surface characteristics on vehicle fuel consumption can be considered reliable and ready to be implemented in the LCA analysis and maintenance programming of road pavements, to support decision making at strategic level