Instituto Tecnológico y de Estudios Superiores de Monterrey

(1)

Instituto Tecnológico y de Estudios Superiores de Monterrey

Campus Monterrey

School of Engineering and Sciences

Real fuel consumption in the main logistic corridors of Colombia

A thesis presented by

Oscar Sebastián Serrano Guevara

Submitted to the

School of Engineering and Sciences

in partial fulfillment of the requirements for the degree of Master of Science

In

Engineering Science

Monterrey Nuevo León, June 15^th, 2021

(2)

ii

Dedication

To my family, my parents Oscar Remigio and Jhoanna Marlene, whom I love with all my soul; my brother, Christian, who has always been my example of perseverance, discipline, and personality.

To all the people who have shown their support during my academic preparation, I hope God and life reward you generously.

To my advisor, PhD. José Ignacio Huertas, that without his knowledge, support, and supervision, the development of this work would not have been possible.

To my Mexican, Colombian, Argentine and Ecuadorian friends, with whom I have shared the best two years of my life, a young Ecuadorian man loves you.

To God and María because I have always felt them with me.

This work has been carried out with a lot of affection, patience, and love, thanks to what I have received from you.

Oscar Sebastián.

(3)

iii

Acknowledgements

Thank you!

To my parents for the support, education, and love. I am nothing without you.

To my brother for being my best friend all my life.

To my friends, I do not name them because I am afraid to forget someone, but if you read this you will feel this achievement as yours.

To my advisor, thanks for everything, Doc!

To PhD. Jenny Díaz, thanks for your patience and time to check this work.

To my friends and colleagues from the best research group in the world: Energy and Climate Change Research Group, let's keep doing “cosas chingonas”.

To PhD. Daniel Prato and MSc. Lina Tabares, thanks for your support in the development of this project.

To MSc. Gustavo Álvarez, thanks for recommending me to Dr. Huertas.

To MSc. Jhon Jairo Pabón, thanks for your technical input from transport industry.

To Fernando and Virginia, thanks for the time getting the information.

To Conacyt and Mexican Government, thank you for financing my maintenance in your beautiful country and for supporting science.

To Tecnológico de Monterrey for the tuition scholarship, the best people in the world and the best two years of my life, I hope there will be many more!

Thank you all!

Oscar Sebastián.

(4)

iv

Real fuel consumption in the main logistic corridors of Colombia

by

Oscar Sebastián Serrano Guevara

Abstract

Fleet managers state that fuel consumption accounts for 50% of the operating costs of their cargo and passenger vehicle fleets. Aiming to reduce these costs, transport companies contract telematics services to track their units and monitor fuel consumption along with other variables. The gathered information is used to alert managers on events like excessive fuel consumption, abrupt breaks and needs of mechanical maintenance. We propose the use of this information to determine the fuel consumption of cargo vehicles at each km of the main roads of a given region and the influence of altitude, road grade and vehicle age on it. As a case study we studied the fuel consumption in the main logistic corridor of Colombia which are characterized by having a highly variable topography. Toward that end, we compared the fuel consumption monitored by a telematic system on 46 vehicles of different cargo capacity with the estimated by an energy balance model and observed that they are highly correlated (R²>0.99). Then, we used the calibrated model to obtain the km-by-km fuel consumption. This information is used by authorities to obtain a close estimation of the cost of cargo transport, the greenhouse gases emissions, and to identify locations with unusual high fuel consumption. Furthermore, the slope of the linear correlation (Cf) decouples the fuel consumption associated to driving style (human factors) from other influencing factors. Then, we observed that the effects of altitude and vehicle age on fuel consumption are negligible and that most of the vehicle technologies studied has not improved in practice their real energy performance during the last 20 years.

Keywords: Freight transport, Specific fuel consumption, Telematics, Vehicle energy efficiency

(5)

v

List of Figures

Figure 1 Illustration of the methodology followed in this work. ... 5 Figure 2 Region studied in this work (IGAC, 2012) ... 6 Figure 3 Physical forces acting on vehicles on a positive road grade. ... 12 Figure 4 a.) Average cargo-vehicle speed measured at each km of the main Colombian logistic corridors. b.) Speed-acceleration frequency diagrams for each logistic corridor. c.) operating modes frequency diagram. d.) driving cycle ... 16 Figure 5 Results calibrating the fuel consumption model. a.) Evolution of the measured fuel consumption contrasted with variations in road altitude for a 3S3, 52 ton truck traveling on corridor 2 (Bva - Bta). b.) Results of the correlation analysis carried out between experimental and analytical results of cumulative fuel consumption for the same case. Frequency distribution of c.) the coefficient of determination (R2) and d.) calibration factor (Cf) obtained 30 trips monitored on corridor 2. ... 20 Figure 6 Km-by-km fuel consumption and CO2 emissions in the main Colombian logistic corridors. a.) specific fuel consumption (SFC) in l/100 t km. b.) CO2 emissions, in g/100 t km. ... 21 Figure 7 SFC a) lognormal distribution, b) distribution by speed bins, c) lognormal distribution fit at a given speed bin and d) dispersion at different VSP and speed ranges. ... 23 Figure 8 Effect of altitude and road grade on fuel consumption for a 2S2 and a 3S3 truck. a.) SFC from vehicle perspective as function of altitude. b.) SFC from logistic perspective as function of altitude . c.) SFC from vehicle perspective as function of road grade. d.) SFC from logisitic perspective as function of road grade ... 24 Figure 9 Variation of Cf and SFC according to specific routes. a) Cf, b)SFC [l/100 km], c) SFC[l/100 t km] ... 25 Figure 10 Effect of vehicle technology and vehicle age on fuel consumption. Variation of a.) SFC [l/100 km] with vehicle age, b.) SFC [l/100 km] with thousands of km traveled, c.) SFC [l/100 t km] with vehicle age, d.) SFC [l/100 t km] with thousands of km travel ... 27

(6)

vi

List of Tables

Table 1. Literature review about telematics system use in transportation. ... 3

Table 2 Characteristics of logistic corridors considered in this work. ... 7

Table 3 Technical characteristics of the vehicles used in this study. ... 8

Table 4 Technical specifications of the instruments used in this work. ...10

Table 5 Characteristic parameters (CP’s) that describe driving patterns of cargo drivers in the main logistic corridors of Colombia. For each CP and SFC, values highlighted the in blue and red correspond to the best worst condition, respectively. ...17

Table 6 Linear regression model p-value for deterioration analyzed technologies. ...27

(7)

vii

Introduction

1.1 Overview

At the aggregate level, it is estimated that fuel consumption is responsible for about 50% of the operating cost of road transportation (Gohari et al., 2018). However, the fuel consumption of vehicles, km-by-km, on the main freight transport routes is unknown, especially in countries with high mountain topography. This information is relevant for freight vehicle fleet managers who are looking to reduce their operating costs. It is also pertinent for government authorities that seek to improve infrastructure conditions that reduce the logistics costs of transporting products and therefore increase the country's productivity. Furthermore, it is also relevant for environmental authorities who keep track of the emission of greenhouse gases of the transport sector in the process of demonstrating the accomplishment of the country´s reduction goals.

Road transport contributes to approximately 75% of the freight transport in China and Europe (Lv et al., 2019; Wiegmans et al., 2018). Although it only represents between 5% and 8% of the current vehicle fleet (BaptistaVentura et al., 2021), it is responsible for more than 40%, 33%, and 50% of the CO2, NOx and PM2.5 worldwide emissions, respectively. In addition, the demand for this type of transport is constantly growing and it is expected to triple by 2050 (IEA, 2017).

The real fuel consumption as well as the tailpipe emissions depend mainly on factors associated with the vehicle technology, the level of maintenance applied, the fuel characteristics, driver factors and external factors such as topography, state of roads or traffic, which are not considered when manufacturers measure such values following standard laboratory procedures (Giraldo & Huertas, 2019).

At a regional scale, transport cost depends on external factors such as road grade, traffic, weather conditions (Bousonville et al., 2019). Carriers consider that road grades and weather conditions have a high effect on fuel consumption and therefore the cost of operating in regions with high temperatures and/or steep road grades is considerably higher than in countries with predominantly flat regions during cold seasons. Seeking to increase the competitiveness of their countries, governments propose the integration of logistics corridors to reduce the logistics costs of freight transport and its environmental

(9)

2

impact, in addition to contributing to the development of multimodality in transport systems and reducing traffic congestion (Mintransporte, 2021b; Stefanović et al., 2020).

On the other hand, the authorities that regulate transport activity argue that obsolete technologies and vehicles´ age are the factors that have the greatest impact on fuel consumption. However, besides the use of old technologies, the high operating cost faced by transportation companies is attributed to improper driving practices, and traffic congestion (Eswar et al., 2018). These factors result in an increase of the fuel consumption that represents an additional cost between 30% and 50% of their operating costs. (Bracco et al., 2016; Díaz-Ramírez et al., 2018). In addition, most of these companies face intentional fuel and vehicle losses. Finally, freight transport companies are aware of the high impact on their operative cost of broken vehicles in the middle of the road due to lack of proper maintenance.

As a solution to some of these problems, more than 20 years ago, companies started to implement GPS in their units and to monitor the real-time position of their vehicles. With the fuel injection technology, these companies expanded their monitoring system to include measuring fuel consumption and other engine operating variables. These systems focused on i) vehicle location; ii.) inappropriate driving event alert messages, such as abrupt breaking events; and iii.) predictive maintenance alerts. These systems are known as telematics systems. They combine sensor technologies, telecommunications, cloud services, and data analytics to provide roadside assistance to drivers and remote diagnosis to fleet managers.

1.2 Literature review

New uses have emerged for the information gathered by telematics systems (Table 1).

(Léonardi & Baumgartner, 2004) showed that vehicle telematics and computerized routing play an important role in reducing fuel-related costs and emissions of trucking companies. The analysis of these data opens the possibility of using telematics information for routing (Rodrigues et al., 2018), effective delivery scheduling, urban planning, and policymaking (Laranjeiro et al., 2019). Fleet managers use them to improve fuel economy, safety and maximize vehicle utilization (Perrotta et al., 2019).

(10)

3

The literature reports that data obtained from telematics systems can be used to obtain transportation efficiency metrics, such as fuel consumption, idle use, driving modes, as well as management and logistics.

Table 1. Literature review about telematics system use in transportation.

Authors Vehicle Fuel Modelling Validation Telematics application Madhusudhanan

et al., (2020)

HDV Diesel and CNG

Energy demand model

In-service driving cycles

Efficiency metrics (emissions, energy consumption) / t km Farzaneh et al.,

(2020)

HDV Diesel and CNG

- - Idling metrics: duration,

emissions, used fuel Bousonville et al.,

(2019)

MDV HDV

- - - Predict fuel

consumption using machine learning Smirnov &

Picalev, (2019)

LDV - Energy demand

model

Standard urban driving cycle

Evaluate and analyze fuel consumption Kwan & Boodlal,

(2014)

HDV Diesel - - Identify unsafe events,

eco-driving Ayyildiz et al.,

(2017)

HDV LCV

Diesel Energy demand model

Standard urban driving

Reduce fuel

consumption average 6% for HDV and 1% for LDV

Schröder &

Cabral, (2019)

HDV Diesel COOPERT III

model for estimate specific fuel consumption (SFC) based on speed.

Telematics data

Route optimization for vehicles that run fixed routes.

Díaz-Ramírez et al., (2017)

HDV MDV

Diesel - - Fuel efficiency related

variables analysis and eco-driving.

Nevland,

Gingerich, & Park, (2020)

HDV - - - Classify parking

locations for HDV.

Paffumi, De Gennaro, &

Martini, (2018)

- - - - Mobility analysis; pilot

study of travel behavior in Europe.

1.3 Motivation

The reports outlined in Table 1 show that the information obtained from the telematics systems in vehicles has not been used to describe or evaluate actual km-by-km fuel

(11)

4

consumption nor to evaluate the effect of external variables such as vehicle technology, vehicle age, road grade, load capacity, and altitude on fuel consumption. To close this gap, this work focuses on determining the actual fuel consumption km-by-km and CO2

emissions of freight vehicles in the main logistics corridors in Colombia, and then, to use this information to assess the impact of the aforementioned variables.

1.4 Objectives

In the process of pursuing these goals, we developed the following contribution to new knowledge.

• A method to determine the fuel consumption of vehicles km-by-km based on data collected by telematics systems.

• A method to separate the human factor from other factors that influence fuel consumption.

• A quantification of the effect of a vehicle's age on fuel consumption.

• An estimate of the effects of altitude and road grade on fuel consumption.

(12)

5

Materials and methods

Figure 1. illustrates the general methodology for determining the actual fuel consumption, km-by-km, of freight vehicles that circulate on the main logistics corridors of a given region. These vehicles are of different ages, technologies, and sizes, which use a telematic system to monitor their normal operation for an extended period of time.

In a data analysis phase, which involves the development of an energy balance model, driving patterns identification, and statistical analysis, we finally report for each of the selected corridors, the fuel consumption at each km of the road, and evaluate the effect that the altitude, road grade, and vehicle age have on fuel consumption.

Figure 1 Illustration of the methodology followed in this work.

2.1 Study Region

We selected Colombia because the road transport is essential for its economy and it has a road network covering a region with high mountains, which is a characteristic of most of the Latin-American countries. Its main corridors cross the Andes mountains connecting the main states and municipalities of the country. Its road network consists of about 205,000 kms of road, of which 18,000 kms correspond to primary roads (highways). Additionally, It has roads in regions located over 3800 meters above sea level (masl) and with road grades of up to 13%. Figure 2 shows the topography and the

-5 -2.5

0 2.55 0%

2%

4%

6%

8%

10%

0 1530 45 60 75 90

Freight transport vehicles Different age, size, technologies

• Driving pattern

• Fuel consumption Main logistic corridors in the country

Data analysis

evaluation

• Speed

• Altitude

• Road grade

• others

Monitoring by telematics Study region

(13)

6

road network of the selected study region. More than 70% of freight transport is carried out on road in Colombia. Between 2016 and 2020, more than 100 million tons were transported per year (Mintransporte, 2021a). The country's economy also has a high dependence on fossil fuels: 75% of total energy consumption (Hannah & Max, 2017) of which 40% corresponds to the transport sector and a third to the transport of goods (MinMinas & UPME, 2016).

Figure 2 Region studied in this work (IGAC, 2012)

2.2 Selected routes

For this work, freight transport trips occurring along the different logistics corridors were considered. Logistics corridors refer to the physical means that facilitate the exchange and development of trade in the country, through which the cargo of both foreign trade and domestic trade is mobilized. A logistics corridor integrally articulates one or several origins and destinations in aspects such as transport infrastructure, information and

(14)

7

communication flow, business practices, and activities that facilitate trade (MinTransporte, 2014).

The National Highway Institute (INVIAS) uses a code of six digits to identify the roads in the country. The first two digits represents the name of the route (main cities of origin and destination), the next two specify the section of the route (places that it connects) and the last two refer to the length in kilometer of that road. Table 2 summarizes the main characteristics of the eight logistic corridors. Additionally, INVIAS uses references posts located at every km on the main roads of the country whose location and altitude and additional information are available on the website of the Ministry of Transportation.

We assigned to each reference post variables such as average speed and actual fuel consumption of the vehicles crossing by.

Table 2 Characteristics of logistic corridors considered in this work.

ID Logistic

corridor Main cities

Max altitude [masl]

Min altitude [masl]

Max road grade [%]

*Roads length [km]

1 Bogotá Costa Caribe

Bogotá, Puerto Boyacá, Ciénega, Barranquilla, Cartagena

2840 0 8.7 2225

2 Bogotá Buenaventura

Buenaventura, Cali, Bogotá, Popayán, Pasto, Ipiales

3246 0 13.5 1617

3 Bogotá Cúcuta Bogotá, Tunja,

Bucaramanga, Cúcuta 3850 70 6.6 1477

4 Medellín Costa Caribe

Medellín, Cartagena,

Montería, Cali, Manizales 3672 0 7.4 2446

5 Medellín Bucaramanga

Medellín,

Barrancabermeja, Bucaramanga

1546 67 5.3 450

6 Bogotá Villavicencio

Bogotá, Duitama, Yopal,

Aguazul, Villavicencio 3465 220 6.7 831

7 Bogotá Orinoquía

Arauca, Saravena, Yopal, Villavicencio, San José del Guaviare

645 120 2.5 1218

8 Bogotá Putumayo

Bogotá, Puerto Asís,

Mocoa, Neiva 2785 260 6.7 793

* Considered as the sum of the length of road sections in corridors.

2.3 Vehicles

(15)

8

Table 3 describes the 46-vehicle fleet monitored in this work. All of them are class 8 (gross vehicle weight greater than 14.96 tons). According to the classification of the Ministry of Transportation (Mintransporte, 2004) they are 2S2 and 3S3. The first digit denotes the number of axles of the truck, while the "S" specifies that the vehicle has a semi-trailer, and the last digit states the number of axles on the semi-trailer. Three vehicles were more than 10 years old, 34 between 5 and 10, and five were less than 5 years old. This information could not be identified for the remaining two. Seven vehicles comply with the Euro IV and Euro V emission standards while the other 37 vehicles comply with the EPA 98 standard. In this country, until 2015, cargo transport vehicles that entered the country must comply with this EPA 98 standard (MinTransporte &

Carga, 2019). The listed vehicles are identified as MFxx based on make, model and model year, this for further analysis.

Table 3 Technical characteristics of the vehicles used in this study.

Id

Curb vehicle weight [t]

Manufacturer and vehicle model

Model year

Numbe r of vehicle s

No.

trips

Configuration – gross vehicle weight [t]

Emissions control technology

Transmission ratios*

(16)

9

MF1 6.6 Kenworth T3 2011 3 45

2S2 – 32

EPA98

9-speed transmission 12.64/8.81/6.55/

4.77/3.55/2.48/

1.85/1.34/1/5.25 MF2

7.2

International

Workstar 7600 2012 1 33

MF3 International

Workstar 7600 2013 2 58

MF4 International

Workstar 7600 2020 1 29 EuroV

MF5

7.8

Kenworth T8 2007 2 62

3S3 – 52

EPA98

18-speed transmission 14.40/12.29/8.86/

7.30/6.05/5.16/

4.38/3.74/3.20/

2.73/2.29/1.95/

1.62/1.38/1.17/

1.00/0.86/0.73/

4.30

MF6 Kenworth T8 2012 7 200

EuroV

MF11 7.5

International

9400i 2012 6 13

EPA98

MF12 International

9400i 2013 2 10

MF13 7.8

International

9900i 2008 1 28

MF14 International

9900i 2012 1 42

MF15 6.5 International

Prostar+122 2018 1 28 EuroIV

MF16 6.8 Freightliner

Columbia 120 2013 1 21

EPA98 MF17 9.0 Freightliner

Columbia 120 2015 1 36

Cascadia 125 2012 1 35

New Cascadia 2020 1 1 EuroV

2.4 Data gathering

The position, speed, and fuel consumption of the specified vehicles were monitored as follows (Table 4):

(17)

10

Position and speed: This information is reported by the telematics system from the readings of a GPS. The vehicle speed is given in km/h and position in decimal degrees of longitude and latitude. These values are reported simultaneously with an average frequency of 1 data every 20 seconds.

Altitude: The telematics system does not report the vehicle altitude. It was obtained through the Digital Elevation Model (DEM) of GPS Visualizer tool (www.gpsvisualizer.com/elevation) using as input the location of the vehicle. When the vehicle crossed nearby a reference post, altitude was obtained as the altitude of the reference post which is reported by INVIAS. A correlation analysis between these two methods showed that they highly correlated (R²= 0.99) and report similar values (slope=

0.99)

Fuel consumption: The telematic system reports from time to time the engine operative variables. One of them is the accumulated fuel consumption which is obtained from the engine control unit (ECU) at very low frequency (~1 data every 20 minutes). Fuel consumption is determined through the injection time or through mass air flow and the air-fuel ratio. This last method was previously validated by (Pepper, 2010) who obtained an error lower than 3%. The specific fuel consumption was computed as l/100 km and l / 100 t km. That is the fuel consumed from the vehicle perspective (l/100 km) and then divided by the reported transported load in tons (t). This metric is of great interest from the logistics perspective (Díaz-Ramirez et al., 2017).

Vehicle weight: the curb vehicle weights, and the load weight of each trip monitored trip, were obtained from the National Registry of Cargo Dispatch of the Ministry of Transportation. This office keeps record of every single solid and liquid cargo trip carry out in Colombia using its main logistic corridors.

Table 4 Technical specifications of the instruments used in this work.

Variable Instrument / application Technical characteristics

Position, speed GO9/Geotab Position: ~2m

Speed: ~0.05 m/s

(18)

11 Accumulated fuel

consumption [l]

Not constant frequency. Reported by the engine control unit (ECU) through the 9-pin diagnostic port.

Altitude Digital Elevation Model of GPS visualizer

Based on (NASA JPL, 2013) Resolution: ~90m

Transported load Weighing machines

Comply with resolution 4100 12/28/2004 by (Mintransporte, 2004) Certified by National Accreditation Organism (ONAC)

2.5 Monitoring campaign

Vehicle monitoring was carried out through the telematic system run by an external partner (Navisaf). They use a GO9 device (Table 4) and provide the monitoring services to the largest cargo fleets in Colombia. This study uses the information collected from 839 trips (> 3.5M data points) made between January 2018 and June 2020 by the 46 vehicles described previously.

2.6 Data Analysis

Driving patterns: From the collected speed and time data, the driving pattern of the monitored vehicles in the region of interest was determined. Driving patterns refers to the way in which vehicles are driven in a given region. It is associated to human aspects rather than external factors such as vehicle technology, road conditions, traffic etc.

Clearly driving patterns are affected by these external factors. Driving patterns can be represented through i.) a set of characteristic parameters (CPs), ii.) a representative driving cycle, iii.) a diagram of speed and acceleration frequency distribution (SAFD), or through iv.) a vehicle-specific power frequency (VSP) diagram, which is the power required by the vehicle divided by its mass (Mahesh et al., 2019). We obtained those representations of driving patterns for each of the roads considered. The representative driving cycle was obtained following the fuel based micro trip method proposed by Quirama et al. (2020).

Fuel consumption determination: As mentioned above, the accumulated fuel consumption is reported by the telematic system at lower frequency than speed and

(19)

12

location. This fact limits the use of telematic systems to evaluate fuel consumption using existing methodologies. Then, aiming to obtain average fuel consumption at every km of the main logistic corridors, two actions were carried out: i.) The vehicle speed data was interpolated using a linear function to obtain speed data at a common frequency. In practice researchers use 1 Hz. This linear interpolation means that between two experimental points the vehicle is assumed to move at constant acceleration. ii.) A model to obtain the 1 Hz fuel consumption was developed. Next, we describe said model.

The energy consumed by a vehicle is the required to overcome the forces that restrict the free motion of the vehicle. Those forces are the Rolling resistance (Fr), Drag (Fd), gravity (Fg), and inertial forces which include the mass of whole vehicle (M) and the equivalent mass of the rotating masses (Mfi). This last mass depend on the in use (i) transmission ratio (NTDi in Table 3) (Gillespie, 1992; Peng, Liu, Parnell, & Kessissoglou, 2019). These forces are illustrated in Figure 3.

Figure 3 Physical forces acting on vehicles on a positive road grade.

The power delivered by the engine (Pe) is the sum of the power associated to each of these forces (Fj V) divided by the mechanical efficiency of the power train (𝜂𝑡).

𝑃𝑒 = (𝐹𝑑+ 𝐹𝑟+ 𝐹𝑔 + 𝑀𝑀𝑓𝑖 𝑎 ) 𝑉

𝜂_𝑚 = 𝑣𝑓̇ 𝜌𝑓 𝐿𝐻𝑉 𝜂𝑡ℎ (3) 𝐹_𝑑 =1

2 𝑐^𝑑𝜌_𝑎𝐴_𝑓𝑉² (4) 𝐹_𝑟 = 𝑓_𝑟𝑀𝑔 𝑐𝑜𝑠 𝜃 (5) 𝐹_𝑔 = 𝑀 𝑠𝑖𝑛 𝜃 (6) F_g

(20)

13

𝑀𝑓𝑖 = 1 + 0.04 𝑁_𝑇𝐷𝑖+ 0.0025 𝑁_𝑇𝐷𝑖² (7) 𝜂_𝑡ℎ = 1 − (1

𝑟)^𝑘−1 𝛼𝛽^𝑘− 1

𝑘𝛼(𝛽 − 1) + 𝛼 − 1 (8) The power delivered by the engine is supplied by the chemical energy contained in the fuel consumed (right hand side of Equation 3). It is calculated as the product of the volumetric flow rate (𝑣_𝑓̇ ), density (𝜌_𝑓), and the low heating value (LHV) of the fuel with the engine thermal efficiency (𝜂𝑡ℎ). This last efficiency depends on the engine compression ratio (r) and can be estimated using Equation 8 as a first approximation.

We used a pressure ratio of 𝛼 =2 and a volumetric cutoff ratio of 𝛽=2.5.

The model outlined above does not include additional loads associated to the operation of the vehicle like air conditioning, lighting, engine ventilation and so on. Furthermore, this model needs to be calibrated to include the effects of the mechanical efficiencies of the power train components.

Based on the 1 Hz speed data, we used this model to calculate the fuel consumed every second and compared the cumulated values with the reported by the telematic system. A linear regression analysis was performed for each vehicle and each trip.

Coefficient of determination (R²) near to one will indicate that the model predicts precise values of fuel consumption and that it grasps the main factors influencing fuel consumption. The slopes (Cf) of the liner correlation evaluate the accuracy of the model.

We expect that Cf will be different than one and we will use it as a calibration factor.

Then, the calibrated model was used to calculate the 1 Hz fuel consumption.

Based on those results, we calculated the fuel consumed by each vehicle each time it crossed a given reference post. Within the calculation we included data from ~100 m before and after the exact location of the reference post. We extended this distance up to 500 m and observed no relevant differences.

Results were reported as the specific fuel consumption (SFC) expressed in terms of liters of fuel per 100 km traveled and total vehicle weight (l/100 t-km). We selected this units aiming to compare the performance of vehicles of different sizes. Since payload can have a large range of variation (0-34 t), we used the gross vehicle weight (GVWR) instead of the curb vehicle weight (weight the empty vehicle). Díaz-Ramírez et al (2017) showed that when using this units, the SFC of vehicles exhibit a log normal distribution.

(21)

14

We repeated the process each time the vehicle crossed by the same reference post.

We remain that two-way roads have the same reference post at each way of the road.

Therefore, there is a need to keep track of the traveling direction of the vehicles for future calculation. Then we observed the frequency distribution of SFC at each reference post and hypothesized that it should be of the same shape for all vehicles regardless of the total weigh and vehicle technology. Under that assumption we reported the mean average value of SFC for each reference post which are located at every km of the corridors considered in this study.

Effect of altitude on SFC: We will show later that our model estimates cumulative fuel consumptions highly correlated with experimental results reported by the ECU. We highlight that this model computes the energy demanded to engine by the road loads.

These loads result from variations of the vehicle speed. i.e., they result from the driver´s habits when driving or driving style. Certainly, those loads, and the driver style, are influenced by external factors like traffic. This observation means that if they were not any other additional source of energy consumption, the slope of the linear regression between calculated and measured fuel consumption (Cf) should be equal to one. Furthermore, it means that variations of Cf when the vehicle travel at different altitude will reflect and quantify the effect of altitude on fuel consumption. Using this reasoning, we selected 1 km segments of the road with road grades smaller than 0.5%

located at different altitudes and plotted the observed local Cf as a function of altitude.

We repeated the analysis for several vehicles. We will show that altitude does not have a relevant effect on SFC.

Effect of vehicle deterioration on fuel consumption. Following the previous reasoning, we plotted Cf as function of vehicle age and total km traveled (odometer readings) per vehicle technology to observe the effect of vehicle deterioration on SFC. In this case we used a fixed origin-destination points and included as many trips as the same vehicle performed between the same window of time or odometer reading. In all cases, the length of the road segments considered were longer than 100 km. This analysis assume that vehicles are well maintained, which is an acceptable assumption considering that they are operated by companies with large fleet of vehicles.

(22)

15

Effect of vehicle technology on fuel consumption. Following the aforementioned process, we plotted Cf as function of vehicle model year per trademark limiting the analysis for data obtained when the vehicles were less than one year old. This plot shows the year-by-year improvements that trademarks have been including in the vehicles they offer to their clients.

Determination of the CO2 emissions. For the same fuel composition (same carbon to hydrogen ratio) the vehicles CO2 emissions are proportional to their fuel consumption (Fontaras et al., 2017). For the case of current Colombian diesel, for every liter of fuel consumed, 26.4 grams of CO2 are produced. Results will be reported in terms of gr/100 t-km.

Results

3.1 Driving patterns

(23)

16

Figure 4.a shows the km-by-km mean vehicle speed measured in the corridors listed in Table 2, obtained monitoring the normal operation of 46 cargo vehicles between January 2018 and June 2020. Then, Table 5 describes the driving pattern exhibited by the drivers of these vehicles on these corridors by means of a set of characteristic parameters, Figure 4.b by means of the SAPD diagrams, and Figure 4.c by means of VSP diagrams. Finally, Figure 4.d describe those driving patterns by means of the representative driving cycle. Similar plots are presented in Appendix A for each corridor.

a. b.

c. d.

Figure 4 a.) Average cargo-vehicle speed measured at each km of the main Colombian logistic corridors. b.) Speed-acceleration frequency diagrams for each logistic corridor.

c.) operating modes frequency diagram. d.) driving cycle

-5.00 -2.50

0.00 2.50

5.00 0%

2%

4%

6%

8%

0 15 30 45 60 75 90

Frequency

Speed [km/h]

0%

10%

20%

30%

40%

50%

-27 -21 -15 -9 -3 3 9 15 21 27

Frequency

VSP [kW/ton]

high speed medium speed low speed

0 20 40 60 80

0 200 400 600 800 1000 1200 1400

Speed [km/h]

Time [s]

(24)

17

Table 5 Characteristic parameters (CP’s) that describe driving patterns of cargo drivers in the main logistic corridors of Colombia. For each CP and SFC, values highlighted the

in blue and red correspond to the best worst condition, respectively.

Logistic corridor

Characteristic parameters

SFC Max

speed

Average speed

Std. dev.

speed

% of idling time

% of

acceleratin g time

% of

deceleratin g time

% of

cruising time

Km/h Km/h Km/h % % % % l/100 t km

All 114 35.25 20.72 6.78 18.58 19.20 55.43 1.44

1 114 41.34 21.96 4.01 21.06 20.94 53.99 1.13

2 92 31.10 17.86 4.08 15.29 16.28 64.35 1.58

3 101 36.19 20.14 9.35 14.32 15.05 61.29 1.37

4 92 40.23 21.20 3.67 24.22 24.72 47.40 1.47

5 100 46.98 22.25 5.21 20.61 21.05 53.13 0.98

6 101 35.36 19.30 7.57 15.06 14.83 62.55 1.32

7 87 38.66 21.46 5.68 20.27 20.18 53.87 1.42

8 85 36.36 17.40 5.41 16.75 16.00 61.84 1.31

Table 5 shows that that overall average speed of these corridors is ~ 35.3 km/h, which is low compared with the average values reported for US (99 km/h). It also shows that corridors 5 (Mllin-Bmga) and 2 (Bta-Bvra) are the fastest and slowest corridors, respectively. However, previous comparisons are biased because the roads topographies are highly unsimilar. For example, corridor 5 goes from 67 to 1546 masl in 450 km with an average road grade of 0.11% and average positive road grade of 3.34%

crossing a single mountain chains, while corridor 2 goes from 0 to 3242 masl in 1617 km with an average road grade of 1.04% and average positive road grade of 8.34%, crossing 3 mountain chains. The maximum allowed road grade is 7% but corridor 5 exhibit sections with up to 13% (Table 2). Despite being the slowest corridor, it is the most used in Colombia with 1.018.291 trips per year. For this reason, the Colombian government started 18 years ago a project to speed up this corridor by means a series of tunnel that cross these mountains. Just until today, it started to operate and contain the largest tunnel of the country with 8.65 km long (MinTransporte, 2020).The information provided in this work can be used to identify the sections of the corridor that require similar type of modernization work.

(25)

18

Figures 4.b-c shows that these heavy-duty vehicles are operated with long lasting periods of idling. We double checked that data with the engine off were excluded from the analysis. These large idling times are due to stop and go areas on the primary roads due to infrastructure improvement works, this areas are very common in our study region. Additionally, these large idling times occurs because drivers sleep within the cabin with the air conditioning system on and therefore, they should keep the vehicle engine on. Currently there are electric technologies that addresses this need of providing air conditioning for the vehicle cabin without using the vehicle engine (Koç et al., 2016). Additional work is required to evaluate the economic feasibility and environmental impact of these technologies for the case of the logistic corridors studied in this work.

Finally, Figure 4.d shows the representative driving cycle exhibited by cargo drivers in the Colombian roads. Since it contains the loads on the vehicle, it can be used for the optimization of the vehicles power train configuration or any strategy of energy consumption in this kind of vehicles such as for example the operation with natural gas eighter as compressed gas or in liquified state. Appendix A shows the second-by- second speed series.

3.2 Fuel consumption

For illustrative purposes, Figure 5.a shows the measured cumulative fuel consumption along corridor 2. It also illustrates the corridor topography and shows the drastic fuel consumption increase when the vehicle traves uphill. Figure 5.b compares it with the obtained analytically and shows the high correlation (R²>0.98) between measured and calculated values of fuel consumption. It also shows that the calibration factor or slope of the linear regression (Cf=1.09) is near one, indicating that the model captures the most influencing factors on fuel consumption. This figure also shows that the model started the cumulative fuel consumption in zero while the measured value started at the offset of the linear regression (104255.4 L). This offset will not be considered in future analysis.

(26)

19

We repeated the analysis for all 839 trips monitored of the 46 vehicles and observed in all cases high correlations (R²>0.90, Figure 5.c), especially when the trips are carried out on flat roads (R²>0.97). However, when the trips occur downhill that correlation reduce to an average R²~0.94 and when they occur uphill reduce furthermore to an average R² ~0.92, indicating that under those circumstances additional influencing factors become slightly more relevant. For example, when these heavy-duty trucks at full load go uphill the engine demand additional cooling loads that were not considered in the model.

Similarly, Figure 5.d shows the frequency distribution of the calibration factors. It shows that when the vehicle travels on flat roads the calibration factors are grouped in the range of 1-2 exhibiting a lognormal distribution according to Anderson-Darling test with 5% level of significance (AD=1.95). However, when the vehicles travel uphill, or downhill additional factors show up, dispersing the distributions of R². Thus, for future comparative analysis is preferable to consider only tracks of the road with negligible road grade (<0.1%).

a. b.

c. d.

0 50 100 150 200 250 300 350 400 450

0 500 1000 1500 2000 2500 3000 3500

0 200 400

Real accumulated fuel [l]

Altitude [masl]

Distance [km]

Altitude profile Cumulative fuel

consumption y = 1.09x + 104255.44

R² = 0.98

104200 104300 104400 104500 104600 104700 104800

0 100 200 300 400

Real accumulated fuel [l]

Estimated accumulated fuel [l]

0%

20%

40%

60%

80%

0.90 0.94 0.98

Frequency

R²

Uphill Flat Downhill

0%

20%

40%

60%

80%

1.0 2.0 3.0 4.0 5.0

Frequency

C_f

Uphill Flat Downhill

(27)

20

Figure 5 Results calibrating the fuel consumption model. a.) Evolution of the measured fuel consumption contrasted with variations in road altitude for a 3S3, 52 ton truck traveling on corridor 2 (Bva - Bta). b.) Results of the correlation analysis carried out between experimental and analytical results of cumulative fuel consumption for the same case. Frequency distribution of c.) the coefficient of determination (R2) and d.)

calibration factor (Cf) obtained 30 trips monitored on corridor 2.

3.2.1 Fuel consumption in the main logistic corridors

Using the calibrated model, the actual specific fuel consumption (SFC), expressed in l / 100 t km, was obtained at each reference post (i.e. km-by km), considering the 46 freight transport vehicle, and the 839 monitored trips. Using a geographic information software (Arc GIS), the obtained average values were plotted in Figure 6.a. Adding these values per corridor, Table 5 shows that the fasted corridor (5, Mllin-Bmga) exhibits the least fuel consumption (0.98) while corridor 2 (Bta-Btra) the largest (1.58).

Figure 6.a also shows the location with excessive fuel consumption. It is out of the scope of this work to explain those peaks of fuel consumption. However, we used web map services to visit those locations and explored unsuccessfully potential reasons for those unexpected high fuel consumptions. Qualitatively they are correlated to places with low average speeds. However, we did not find evidence of high traffic on those places. Possibly, those locations correspond to preferred places for drivers to rest or have meals. Possibly, those locations correspond to places where frequently there are intentional fuel losses.

a. b.

(28)

21

Figure 6 Km-by-km fuel consumption and CO2 emissions in the main Colombian logistic corridors. a.) specific fuel consumption (SFC) in l/100 t km. b.) CO2 emissions, in g/100

t km.

Even though the use of an average value of SFC is customary, it could be misleading because i.) the mean value could not be the best descriptor of fuel consumption and ii.) it could depend on other highly influencing factors, such as vehicle size, altitude, vehicle technology and age.

Thus, we started by exploring the SFC distribution of a single vehicle at a given reference post. Figure 7.a illustrates the type of distributions obtained. We applied the goodness-of-fit test and found that a 3-parameter lognormal distribution is the function that best fit this distribution. To confirm that this distribution function fits the experimental data we applied the Anderson-Darling goodness-of-fit test. The critical value of the Anderson-Darling test for a 5% of level of significance is ADCRIT = 2.501 (Jäntschi & Bolboacă, 2018). Hence, the goodness-of-fit test is accepted when the AD statistic is less than this value. We repeated the process for 13 randomly selected cases with 2S2 and 3S3 vehicles and obtained in all cases AD<1.86, confirming that SFC distribution can be described by lognormal distributions. This result implies that the median value of the lognormal distribution is the best metric to describe SFC.

(29)

22

Then, we examined the SFC of each vehicle as function of speed (Figure 7.b) and VSP (Figure 7.c). Again, we explored the SFC at each speed or VSP bin and observed it conserved that it exhibits a lognormal distribution. Figure 7.c illustrates this observation.

The scientific community express fuel consumption and emissions in this way in an attempt to isolate the performance of the engine from the way the vehicle is used (driving pattern or driving habits) (Hart et al., 2002). It means that multiplying median values of 7.d by the corresponding mean values of 4.c, the weighting average value of SFC is obtained (𝑆𝐹𝐶̅̅̅̅̅ = ∑ 𝑆𝐹𝐶𝑖,𝑗 𝑓𝑖,𝑗 ). While 𝑆𝐹𝐶𝑖,𝑗 is characteristic of the vehicle, 𝑓𝑖,𝑗

depends on human behavior. A similar approach has been implemented for quantifying tailpipe emissions in tools for mobile source emission inventories such as MOVES (Hart et al., 2002).

In an additional work, authors are confirming that SFC vs. VSP results depends exclusively on engine technology. In this work we propose a different approach to isolate the performance of the engine from human factors. As described in the methodology section, the calibration factor obtained from the linear correlation between the calculated and measured cumulated fuel consumption achieve this objective. Next, we will use the calibration factor to study the effect of technology and age on fuel consumption.

a. b.

c. d.

7 6 5 4 3 2 1 0 20

15

10

5

0

SFC [l/100 t km]

Frequency [%]

a:-2.277 μ:1.457 σ:0.267 p-value: 0.687 AD: 0.456 n: 159 PR: 15608 3S3 truck

95 85 75 65 55 45 35 25 15 5 9 8 7 6 5 4 3 2 1 0

Speed [km/h]

SFC [l/100 t km] 7 2S2 32t trucks

37 3S3 52t trucks 1.77M data points

(30)

23

Figure 7 SFC a) lognormal distribution, b) distribution by speed bins, c) lognormal distribution fit at a given speed bin and d) dispersion at different VSP and speed ranges.

3.2.2 Effect of altitude on fuel consumption

Figure 8 shows the SFC obtained for all vehicles on 1 km road tracks with negligible road grade (<0.1%), located at different altitudes. Figure 8.a shows that the SFC in terms of the vehicle, that is [l/100 km], is between 20 and 40% greater for truck 3S3, however, when considering the load capacity given by the GVW, Figure 8.b shows that in terms of logistic efficiency, the SFC for 2S2 configuration truck is between 20% and 30% greater than 3S3 truck. Furthermore, there is no evidence that SFC is greater or is increased with altitude. This result agrees with analytical models that predict engine performance. However, it contradicts people’s expectations that assume that fuel consumption increases with altitude.

a. b.

c. d.

9 8 7 6 5 4 3 2 1 0 10

8

6

4

2

0

SFC [l/100 t km]

Frequency [%]

3-p lognormal distribution fit Speed: 60 km/h

7 2S2 32t trucks 37 3S3 52t trucks 281k data points

24 6 15 -12 -3 -21 -30 24 6 15 -12 -3 -21 -30 24 6 15 -12 -3 -21 -30 9 8 7 6 5 4 3 2 1 0

SFC [l/100 t km]

low speed medium speed high speed

VSP [kW/ton]

3000 2500 2000 1500 1000 500 3000 2500 2000 1500 1000 500 400 350 300 250 200 150 100 50 0

SFC [l/100 km]

2S2 truck 32t 29k data points 3S3 truck 52t 216k datapoint

Altitude [masl]

3000 2500 2000 1500 1000 500 3000 2500 2000 1500 1000 500 9 8 7 6 5 4 3 2 1 0

SFC [l/100 t km]

2S2 truck 32t 29k data points 3S3 truck 52t 216k datapoints

Altitude [masl]

(31)

24

Figure 8 Effect of altitude and road grade on fuel consumption for a 2S2 and a 3S3 truck. a.) SFC from vehicle perspective as function of altitude. b.) SFC from logistic perspective as function of altitude . c.) SFC from vehicle perspective as function of road

grade. d.) SFC from logisitic perspective as function of road grade

3.2.3 Effect of road grade on fuel consumption

Similarly, Figure 8.d illustrates that the SFC distribution exhibit log-normal distributions when the vehicle travels on road tracks with different road grades. Figure 8.c shows the boxplot of SFC as function of road grade illustrating that the mean and median of the SFC increase with road grade. t-test and non-parametric tests among road grades bins indicated SCF does vary with road grade. A linear correlation analysis indicated that SFC increase directly proportional with road grade. The constant of proportionality is 1.07 for the 2S2 truck and 1.09 for 3S3 truck. When all vehicles are considered a value of 1.08 was obtained for this constant.

3.2.4 Fuel consumption by route.

We considered three specific routes and then analyze the variation of Cf and SFC according to route characteristics, these trips were all done by different 3S3 trucks (GVWR: 52t), Figure 9. shows its variation and confirm the conclusion stated above that SFC increase with road grade.

a. b.

8 6 4 2 0 -2 -4 -6 8 6 4 2 0 -2 -4 -6 400 350 300 250 200 150 100 50 0

SFC [l/100 km]

2S2 truck 32t 29k data poin 3S3 truck 52t 216k data poin

Road grade [%]

8 6 4 2 0 -2 -4 -6 8 6 4 2 0 -2 -4 -6 9 8 7 6 5 4 3 2 1 0

SFC [l/100 t km]

2S2 truck 32t 29k data points 3S3 truck 52t 216k data poin

Road grade [%]

Instituto Tecnológico y de Estudios Superiores de Monterrey