|Thesis abstract: |
The transportation sector accounts for 22% of global primary energy use and 27% of global CO2 emissions. Transport energy demand in developed countries represents the bulk of the world transport energy use with a share of 65%. Nevertheless, the energy used in transport has grown considerably faster in developing countries during the 1971- 2000 period, at an annual rate of 5% compared with 2.1% for developed countries (Price et al., 2008).
The energy consumption of a vehicle directly influences fuel consumption and air pollutant emissions (CO2, NOx, PM10, etc.) with both economic and social effect. It is well known that the fuel cost has a deep impact on the world economy, while the Carbon dioxide (CO2) is the most important components of greenhouse gases (GHG). Nowadays, reducing the vehicles¿ fuel consumption and emissions are two of the main objectives pursued by institutions; the European Commission plan, new cars in Europe may be required by 2015 to meet a strict fleet wide average of 130 grams of carbon dioxide per kilometer driven; the United States is expected to adopt similar CO2 standards. These ambitious goals may be achieved through the use of alternative primary energy sources, the development of more efficient powertrains and a broader adoption of electric vehicles. Most of the automotive research in this field foremost focuses on the optimization of powertrain¿s components, architectures and control strategies. Although these solutions seem promising for the next decades, they will not likely have a big impact in short term. The technology available is not mature enough and the market penetration of alternative vehicles is negligible. Nevertheless human factors and user¿s modes of transportation have a huge impact both in terms of the vehicle energy consumption and of CO2 emissions. Previous researches demonstrate that by improving the driving-style, a driver can reduce the vehicle fuel consumption from 5% up to 40%, while CO2 emissions may be reduced by leveraging the rising interest in public and electric vehicles. This thesis try to take advantages from one of the most pervasive device, the smartphone, to propose several approaches devoted at decreasing fuel consumption and emissions of combustion and electric vehicles by influencing the user behaviour. The main advantage of the approaches proposed in this thesis is that they rely just on inertial measurements, so they can be directly applied and they can produce a real benefit for the user and for the environment without any additional ad-hoc hardware.
In the very first part of the Thesis, we discuss the vehicle energy conversions steps. In particular, we present an analytical parametric model to compute the vehicle consumption from inertial measurements of the vehicle longitudinal dynamic. We therefore show that the overall vehicle fuel consumption and CO2 emissions depend by the vehicle powertrain and by the upstream process efficiencies. For the sake of clarity, an experimental characterization of electric vehicle efficiencies is carried out.
We then analyze the energy estimation achieved with a low sampling-rate dynamic signal. We firstly propose a signal processing methodology to simulate the effect of down-sampling on measured mission profiles. We propose a statistical characterization of the error introduced by the low sampling rate signal using real-data collected by two cars during several days. Experimental results suggest two methodologies to statistically compensate the average and reduce the variance of the energy estimation error.
Since most of the remaining parts of the Thesis rely on smartphone embedded inertial sensor we develop an experimental analysis and evaluation of these sensors measurements. For this aim we carry out an experimental acquisition campaign on urban and extra-urban roads using two vehicles, and comparing four smartphones¿ measurements with an external reference system. Then, we design a signal processing chain for reducing the signals noise and compensating the device orientation. Finally, we propose a data-fusion algorithm to reconstruct the vehicle longitudinal dynamic and to compute the power consumption.
In the last three chapters of the Thesis we propose three innovative application that exploit inertial measurements.
First, we design and develop a system able to assess in real-time the driving-style. The system is fully integrated in a smartphone application, which acquires the signals related to the vehicle dynamics (velocity and acceleration) and computes three power-related indexes. The system provides visual feedbacks to the driver who modifies his behavior accordingly. Finally, to experimentally prove the effectiveness the proposed approach we design and carry out an experimental campaign with five volunteers and an electric car. Experimental results show that the interaction between the driver and the system improves the driving-style and reduces the overall vehicle consumption.
Second, we propose a novel method to estimate in real-time the CO2 emissions related to the user¿s mode of transportation. We design an algorithm to automatically classify the users mode of transportation in eight classes (car, bus, train, subway, motorcycle, walking, biking, still) using inertial information gathered from smartphone sensors. Since we choose a black-box supervised learning approach we develop a working Android application to collect data for training and validating the model and we design an experimental acquisition campaign with ten anonymous volunteers. Experimental results show that the proposed system is able to identify the mode of transportation with an average accuracy greater than 80%.
Third, we propose a methodology to simulate Electric Vehicles (EVs) and Series-Hybrid Electric Vehicles (SHEVs) energy profiles from measured inertial mission profiles. The simulation approach can be applied to several problems. We focus on two examples: obtain reliable EV grid load profiles and determine the optimal sizing for the battery pack and the range-extender of an SHEV. Examples of the simulation approach and of the two applications are provided with real-data.