Autonomous bucket emptying on hauler

(1)

Institutionen för systemteknik

Department of Electrical Engineering

Examensarbete

Autonomous bucket emptying on hauler

Examensarbete utfört i Reglerteknik vid Tekniska högskolan vid Linköpings universitet

av

Anders Bergdahl

LiTH-ISY-EX--11/4498--SE

Linköping 2011

Department of Electrical Engineering Linköpings tekniska högskola

Linköpings universitet Linköpings universitet

(2)

(3)

Autonomous bucket emptying on hauler

Examensarbete utfört i Reglerteknik

vid Tekniska högskolan i Linköping

av

Anders Bergdahl

LiTH-ISY-EX--11/4498--SE

Handledare: Peter Rosander

isy, Linköpings universitet

Jonatan Blom

Volvo CE

Torbjörn Martinsson

Volvo CE

Examinator: Johan Löfberg

isy, Linköpings universitet

(4)

(5)

Avdelning, Institution

Division, Department

Division of Automatic Control Department of Electrical Engineering Linköpings universitet

SE-581 83 Linköping, Sweden

Datum Date 2011-08-30 Språk Language Svenska/Swedish Engelska/English Rapporttyp Report category Licentiatavhandling Examensarbete C-uppsats D-uppsats Övrig rapport

URL för elektronisk version

http://www.control.isy.liu.se http://www.ep.liu.se ISBN — ISRN LiTH-ISY-EX--11/4498--SE

Serietitel och serienummer

Title of series, numbering

ISSN

—

Titel

Title Autonomous bucket emptying on hauler

Författare

Author

Anders Bergdahl

Sammanfattning

Abstract

This thesis proposes a method of loading a hauler with an autonomous wheel loader efficiently and with high productivity in the short loading cycle. The main idea behind the method is to raise the bucket with only the lift until the center of gravity has reached the height that it will have just as the material will start to fall out of the bucket and on to the hauler. When the center of gravity has reached that height both lift and tilt will be used simultaneously to keep the

center of gravity at a constant height. This method is both faster and more

energy efficient than loading a hauler using only one function at a time. The method has been successfully implemented and validated by loading a hauler with three buckets with the autonomous wheel loader that is developed by Volvo CE in Eskilstuna.

Nyckelord

(6)

(7)

Abstract

This thesis proposes a method of loading a hauler with an autonomous wheel loader efficiently and with high productivity in the short loading cycle. The main idea behind the method is to raise the bucket with only the lift until the center of gravity has reached the height that it will have just as the material will start to fall out of the bucket and on to the hauler. When the center of gravity has reached that height both lift and tilt will be used simultaneously to keep the center of gravity at a constant height. This method is both faster and more energy efficient than loading a hauler using only one function at a time.

The method has been successfully implemented and validated by loading a hauler with three buckets with the autonomous wheel loader that is developed by Volvo CE in Eskilstuna.

(8)

(9)

Acknowledgments

I would like to thank Torbjörn Martinsson and Jonatan Blom at Volvo CE and Johan Löfberg and Peter Rosander at ISY for giving me the chance to do my master thesis with the autonomous machine. I would also like to give my thanks to the rest of the Emerging Technology department on Volvo CE in Eskilstuna for helping me with my work.

(10)

(11)

Introduction

The wheel loader is a machine that is used in many industries all over the world. One of the reasons why they are so common is that they are extremely versatile machines. They exist in many different sizes and therefore they can be used in a large variety of applications. Some of these applications consist of repetitive work cycles and these applications (at least on some sites) could be perfect for an autonomous wheel loader. To investigate if it is possible to use autonomous machines on these sites Torbjörn Martinsson at Volvo CE started in 2008 a project with an autonomous wheel loader. The idea was that master thesis students should carry out most of the work.

1.1 Problem formulation

The problem formulation in this thesis can be formulated in one theoretical prob-lem and one practical probprob-lem. The theoretical probprob-lem is the foundation of the thesis, which the practical problem rests upon. The theoretical problem can be formulated in three questions:

• How to empty the bucket onto the hauler?

• What way of emptying the bucket takes the shortest time? • What way of emptying the bucket uses the minimum energy?

The answer to these questions forms the solution and the practical problem will be to implement the solution on the autonomous wheel loader.

1.2 Limitations

Focus is on path planning and control of the bucket (lift and tilt functions), other issues like finding the hauler and calculate the path to the hauler is not a part of this thesis.

(14)

2 Introduction

1.3 Thesis outline

• Chapter 1 is the introduction. • Chapter 2 describes a Volvo L120F.

• Chapter 3 is about the autonomous wheel loader project and the changes made to the machine.

• Chapter 4 is about the theory necessary for understanding the thesis. • Chapter 5 is about finding out How to perform the emptying.

• Chapter 6 is about how the solution from Chapter 5 is implemented on the autonomous wheel loader.

• Chapter 7 presents the results from measurements. • Chapter 8 is the conclusions and future work.

1.4 Definitions

• Latin is a program at Volvo CE that contains drawings and specifics of the wheel loader.

• Work hydraulic - are the lift and the tilt functions on the wheel loader. • The hauler is a Volvo A25

• PI-controller - proportional-integral controller • GUI - graphical user interface

• Lift angle (θlif t) is the angle φlif tin Figure 1.1

• Tilt angle (ϕtilt) is the angle φattin Figure 1.1

(15)

1.4 Definitions 3

• T - torque [N m]

• J - mass moment of inertiakgm2 • r - radius [m]

• θ, ϕ - angles [rad]

• α - angle from buckets edge to the center of gravity [rad], see Figure 1.2 • p - pressure [P a] • q - flowhm3 s i • P - power [W ] • η - efficiency

• Dpump - hydraulic pump displacementm3

• n - angular velocity in revolutions per second • rpm - unit (revolutions per minute)

• i - gear ratio • F - force [N ]

• g - geometry function that relates cylinder length with angle in m

rad on the

autonomous wheel loader (calculated with a look-up table).

(16)

(17)

Chapter 2

The Machine - L120F

In Figure 2.1 a picture of the L120F is shown. It is a mid-size wheel loader and it is a machine of this type that is used in the autonomous machine project. With a bucket the L120F weighs about 21 000 kg, has a tipping load of 12 000 kg and has a 7.1 litre combustion engine that produces 180 kW (245 metric hp). As most other wheel loaders the L120F has articulated steering and uses hydraulics to move the bucket, or any other attachment.[3]

Figure 2.1: A Volvo L120F wheel loader, which is the same model used in the autonomous project ( c Volvo Construction Equipment - All right limited)

The machine in the autonomous project is modified compared to a standard L120F. The modifications consist mostly of additional utilities (sensors and com-puters) so the machine itself is still very close to a standard L120F. The modifica-tions relevant to this thesis are described in Chapter 3. A more complete overview of the autonomous machine can be found in [2].

(18)

6 The Machine - L120F

2.1 Power distribution

A wheel loader works a bit different then e.g. a car does. When the driver pushes the throttle in a wheel loader it does not correspond to torque request, as it would in a car. In a wheel loader it corresponds to an engine speed request, instead of a torque request. The wheel loader then has an engine speed regulator that will keep the engine speed constant even if the torque requirements changes. The engine powers mainly two systems: the hydraulic system and the powertrain, where the hydraulic has the higher priority.

2.2 Hydraulic system

There are many functions on an L120F that is powered by hydraulics, such as work hydraulics (lift and tilt), steering, cooling fan and brakes. The systems are of type LS (load sensing systems) which means that the pressure over the valves (delta pressure) are constant. To power all these functions there are three hydraulic pumps to assure that these functions work properly. If a pump powers more than one function it is a priority between the different functions, below the first listed function has the highest priority.

Pump 1

Pump one is used for filling of the brake accumulator and powers the cooling fan.

Pump 2

Pump two is only used to power the work hydraulics.

Pump 3

Pump three is used for steering and work hydraulics.

The fact that both pump two and three powers the work hydraulics is of great interest in this thesis. When the wheel loader is traveling without turning the displacement from both pump two and three can be used to power the work hy-draulics. When turning, the flow to the work hydraulics is reduced be the amount of flow that goes to the steering.

2.2.1 Priority in the work hydraulics

In the work hydraulics there is a priority order between the lift and the tilt function. This is due to the fact that the same hydraulic system powers both functions and the fact that it is usually higher pressure in the lift cylinder. The delta pressure over the tilt valve will then become higher than normal. This means that there will be a higher flow through the tilt valve (how high is described by (4.15)) than what is intended. The total flow from the pumps is limited and the extra flow to the tilt makes the lift go slower than intended or stop completely. Note that this

(19)

2.3 Engine 7

only occurs when the lift and tilt functions are used simultaneously. This is highly noticeable when driving a wheel loader to load a hauler. The driver first uses the lift function and when the bucket is close enough to the hauler and has reached the right height the driver will start to tilt. If the diver continues to lift during the tilting the lifting speed will be slower and if the driver is not careful with the tilt the lift can stop completely.

2.3 Engine

The L120F has a six cylinder, 7.1 litre diesel engine called Volvo D7E LA E3. The maximum torque is 1065 Nm at 1500 rpm and the maximum power is 245 metric hp at 1700 rpm. The economic working range is 800-1600 rpm.[3]

(20)

(21)

Chapter 3

The Autonomous Wheel

Loader Project

In 2008 Torbjörn Martinsson started the autonomous wheel loader project with the first two thesis workers. Since then the work has continued with mainly thesis workers and now the goal is to develop a fully autonomous wheel loader.

3.1 Previous work

This thesis is the eleventh in the sequence of master thesis works done in this project. What previous thesis workers have accomplished are that they have cre-ated a hardware platform making it possible to control the wheel loader digitally. They have also implemented an industrial computer that has a real time operating system that runs a Simulink model. It is this computer that controls the machine. There is another computer that runs the GUI and does all the larger and more time-consuming calculations. This computer has Windows and the GUI is written in C#.

There has also been thesis works done on the vision system enabling pile and hauler recognition and scanning while moving. The vision-system is now devel-oped further together with Örebro University. An algorithm for bucket filling has already been developed and implemented and now it is time for the loading sequence.

3.1.1 Additional sensor

These are the additional sensors on the wheel loader that are of interest in this thesis.

(22)

10 The Autonomous Wheel Loader Project

Lidar

The LiDAR (Light detection and ranging)-sensor can viewed as the main sensor and is placed on top of the cabin facing forward. A lidar is working in the same way as a radar does, but uses laser instead of radio waves. The output from the sensor in this application is a point cloud. Each point has two values, the distance to it and an intensity value of the reflection.

Angular sensor

Three angular sensors are used, two for the work hydraulics and one for the waist. The sensor that measures the waist and lift angle measures the angles explicitly while the tilt angle is measured implicitly. For the tilt, the actual angle measured is the Ψtilt in Figure 1.1 and together with the lift angle the tilt angle can be

calculated.

GPS, IMU and translation counter

The GPS (global positioning system), IMU (inertial measuring unit) and transla-tion counter is used together with the waist angular sensor and a Kalman-filter to estimate the position of the wheel loader.

3.1.2 Control system

The lift and tilt functions are controlled digitally by the control system. The system is built up with a model-based open loop controller and a PID-controller. To move the bucket the control system use a model of the work hydraulics to calculates a hydraulic flow request, which corresponds to vale positions for both lift and tilt. Feedback from angular sensors and the translations counter are then used in the PID-controller. Up until now, with the exception for the bucket fill, this system has only been used to follow a ramp. Ramps are characterized by constant acceleration then constant velocity or constant velocity and the constant deceleration or a combination of both. That means the control system is now to be used to follow more complex paths. The control system is a large Simulink model that is compiled with Real-time Workshop and then ran in real-time on the autonomous machine.

3.2 Brakes

During the path planning the velocity of the wheel loader is set and the control system for the brakes tries to keep the predetermined velocity. To calculate the velocity ramp that the wheel loader will use to stop in front of the hauler the system will use the current speed and the pre-set braking distance to create the ramp. The brakes are a complex system and are quite hard to control since they are dependent on various external factors, one factor important being the temperature of the hydraulic oil. To minimize the time delay in the brake system the brakes are applied with minimum pressure (no actual braking) long before the hauler.

(23)

3.3 The goals with the autonomous wheel loader 11

3.3 The goals with the autonomous wheel loader

The goal with the project is to build a wheel loader that can do:

• Two hours work at an asphalt plant at 70% productivity compared to a skilled driver.

• One hour loading in the short loading cycle (see Section 4.1) also at 70% productivity compared to a skilled driver.

3.4 Related work

There is one similar project done by the National Institute of Advanced Industrial Science and Technology, in Tsukaba, Japan. They have completed a short loading cycle with a Yamazumi-4 wheel loader, which is smaller than the L120F and they used a very different sensor configuration. For example they have two laser scan-ners one on each side of the bucket that scans laterally, and a stereo-vision system is used for detecting the pile. From the setup (the large distances between wheel loader, hauler and pile) they used, it is possible to conclude that productivity and efficiency were not highly prioritized. The article [6] summarizes their work.

(24)

(25)

Chapter 4

Background Theory

In this chapter the short loading cycle and the wheel loader’s coordinate system will be defined as well as the theory behind the models that are simulated in the Simulink models later on in the thesis.

4.1 Short loading cycle

The emphasis in this thesis is loading in the short loading cycle, also refereed to as the V-cycle or the Y-cycle from the driving pattern (Figure 4.1). Common material used in this setup is single, gravel, et cetera that is loaded on some kind of load receiver. The Algorithm 4.1 defines the short loading cycle in this thesis and it is taken from Reno Filla in [5]. This definition is more extensive than e.g. the one Takashi Tsubouchi uses in [9]. Although the emphasis is on the loading in the short loading cycle the result of this thesis could just as easily been implemented in the load and carry cycle, which basically is a short loading cycle but with two longer stretches between the pile and the hauler.

(26)

14 Background Theory

Algorithm 4.1 Short loading cycle

1. Fill the bucket: The driver fills the bucket by driving into the pile and uses both lift and tilt functions

2. Leaving the pile: The machine starts to move backwards in the V (or Y) pattern towards the reversing point

3. Deceleration: This state is dependent of state four and starts some time before that state. How long before depends on how the driver decides to decelerate (braking or by using the engine).

4. Reversing: When the remaining distance to the hauler is sufficient for the lift hydraulics to achieve the height necessary for emptying the reversing starts. Drivers usually reverse longer distances than necessary to avoid not being able to get the sufficient bucket clearance at the hauler. This state is also called the reversing point.

5. Translate towards the hauler/receiver: The driver drives the wheel loader towards the hauler and completes the V-shape pattern.

6. Empty the bucket: The machine is moving slowly towards the hauler and material is falling out of the bucket and at the same time the bucket is raised and tilted forward.

7. Translate from the hauler/receiver: The bucket is tilted back and the driver drives back to the reversing point while at the same time the bucket is lowered.

8. Decelerating and reversing: Note that this state does not have to be done at the same place as state 3 and 4 since lowering of the bucket is faster then raising.

9. Translate towards the pile: The machine is driven forward to the next entry point in the pile and at the same time the bucket is lowered and tilted so that the edge of the bucket is parallel to the ground.

10. Decelerate at the pile: The driver usually uses the machine’s momentum to drive the bucket into the pile to save time and fuel.

(27)

4.2 Coordinate system 15

Figure 4.1: The short loading cycle as defined by Reno Filla [5] from which also the picture originates. The numbering corresponds with Algorithm 4.1

4.2 Coordinate system

The wheel loader’s coordinate system is defined with the origin in the cabin and has the x-axis straightforward. The y-axis is to the left and the z-axis is upwards making it a positive Cartesian coordinate system.

4.3 Automotive modeling

Inspiration on how to model the powertrain of the wheel loader comes from [4], which has been modified with a torque converter. Since the wheel loader has an engine speed regulator the engine speed is assumed to be constant even though the torque differs with time. The goal with this section is to, given a certain engine speed, simulate how the wheel loader will move from a stand still. A schematics of the powertrain can be viewed in Figure 4.2.

(28)

Figure 4.2: The schematics of the powertrain modell. Input is engine speed and output is wheel speed.

4.3.1 Torque converter

As seen in Figure 4.3 the torque converter has a pump and a turbine and it is the pump that powers the turbine. The pump’s speed (npump) and the turbine’s

speed (nturbine) will roughly determine the torque ratio (µ) as a function of the

ratio between the two. The ratio nturbine

npump is also referred to as the converter’s slip

and is denoted ν. The output from the Simulink model of the converter is the turbine torque Tturbineand is a function of the pump’s torque multiplied with the

torque ratio, as described in (4.1).

Tturbine= µ(ν)Tpump (4.1)

The torque absorbed by the pump (Tpump) is a function of a reference torque called

M P 1000(ν) and the actual engine speed, as shown in (4.2). The M P 1000(ν) is a

function of the slip and is the pumps reference torque. The engine speed used as reference (nref) was 1000 rpm.

Tpump= M P 1000(ν)

npump

nref

2

(4.2) In the Simulink model of the converter both µ(ν) and M P 1000(ν) are look-up tables and the input are the pump speed (npump) as well as the turbine speed

(nturbine).

(29)

4.4 Hydraulics 17

4.3.2 Transmission and final gear

The transmission and final gear is put into one block since the propeller shaft is not modeled, that is i = itransmissionif inal gear. Then the equations for the

transmission and final gear can be expressed as:

Tturbine= 1_iTwheel (4.3)

θturbine= iθwheel (4.4)

˙

θturbine= i ˙θwheel (4.5)

4.3.3 Wheel

The modeling of the wheel is made according to:

Fwheel = m ˙v + Fair+ Froll (4.6)

Froll = mg(cr1+ vcr2) (4.7)

Fair << Froll (4.8)

The constants cr1 and cr2 are the roll resistance and (4.9) can be derived with

Euler’s law.

(Jwheel)¨θwheel= Twheel− rwheelFwheel (4.9)

4.3.4 Complete model

When all the blocks are put together the following equation can be derived, using ˙v = r ˙θ for circular motions:

(Jwheel+ mr2wheel+ iJturbine)¨θwheel= iTturbine− rwheel(Fair+ Froll) (4.10)

Tturbine= µ(ν)M P 1000(ν) _n pump nref 2 (4.11) where npump= nengine.

4.4 Hydraulics

In this thesis it is the hydraulic flow that is used as the key variable that powers the system and therefore these basic equations deserve some attention. The equations in Section 4.4.2 will later be used to model the wheel loader’s lift and tilt functions. A simplified version of the hydraulic system (only the work hydraulics) is shown in figure 4.4 where there are two hydraulic pumps that powers two cylinder that each has a valve to control the flow. The feedback from the load sensing system (LS) is also shown.

4.4.1 Hydraulic losses

To find the most efficient way of emptying the bucket some losses in the hydraulic system has to be calculated. The losses in this section are not all the losses in the

(30)

hydraulic system but these losses will vary a lot depending on how the bucket is emptied on the hauler.

Figure 4.4: A schematics of the work hydraulics system. The flow through a orifice, in this case a valve is equal to:

qlif t = K1

√

ppump− plif t= K1p∆plif t (4.12)

qtilt = K2

√

ppump− ptilt= K2p∆ptilt (4.13)

where K is the proportionality constant, ∆p is the pressure difference before and after the orifice and ppump= pLS+ ∆p and pLS is the load sensing pressure. The

valve is constructed so that when only one function is used with maximum flow then ∆p ≈ 30 bar. D > 6.4uring simultaneous lift and tilt the LS-pressure is equal to highest of the tilt and lift pressure, that is pLS = max(ptilt, plif t). The flow

(31)

4.4 Hydraulics 19

through the valves is then (if plif t> ptilt):

qlif t= Kp∆p (4.14)

qtilt= Kpplif t+ ∆p − ptilt (4.15)

qpump= qlif t+ qtilt (4.16)

and

qpump,max = npumpDpump (4.17)

npump = ηpumpnengine (4.18)

This can be summarized in the following equations:

qmax(nengine) = ηpumpnengineDpump (4.19)

qmax(nengine) = qtilt+ qlif t (4.20)

where npump and nengineare measured in revolutions per second and Dpumpis the

displacement of the pump. When the flow and pressure is known the power losses in the orifices are calculated with the following equations:

Ploss,tilt= qtilt∆ptilt (4.21)

Ploss,lif t= qlif t∆plif t (4.22)

and the extra power losses due to simultaneous lifting and tilting can be expressed as:

∆Ploss= qtilt(plif t− ptilt) (4.23)

which gives the energy losses as the integral of (4.23):

Eloss= tf

Z

ts

∆Plossdt (4.24)

4.4.2 Lift and tilt cylinder

The hydraulic flow to the cylinders can be derived from the angular velocity for each function. This is important since external systems (sensors, control system et cetera) use radians and radians per second, while internally the system uses hydraulic flow. The relations for the lift can be derived from these two equations

vcyl,lif t=

qlif t

Acyl,lif t (4.25)

vcyl,lif t= glif t(θlif t) ˙θlif t (4.26)

and similar for the tilt:

vcyl,tilt = _A_cyl,tiltqtilt (4.27)

(32)

where Acylis the area of the piston (the positive side for the lift and negative side

for the tilt), vcyl is the velocity of the cylinder and g relates the cylinder length

with the angle in the unit_radm . The final expressions for the flow to lift and tilt cylinder as a function of angle and angular velocity then becomes

qlif t= Acyl,lif tglif t(θlif t) ˙θlif t (4.29)

qtilt= Acyl,tiltgtilt(θlif t, ϕtilt) ˙ϕtilt (4.30)

The equation is some times used inversely, describing the angular velocity as a function of position and hydraulic flow.

4.4.3 Steering cylinders

The articulated steering consists of two cylinders that can work both ways (push on both sides of the piston) but during normal driving the cylinders only pushes on the positive side of the piston. The equations for the relation between angular velocity and cylinder velocity are similar to (4.29) and (4.30), the only differences are the cylinder area and the geometry. During normal driving there is only one cylinder pushing at a time. With the coordinate system that is defined in Section 4.2 and if the waist angular velocity is positive, then the cylinder to the right is in use and vise versa.

4.5 Center of gravity

The center of gravity can easily be derived from Figure 4.5 as a function of the lift and tilt angles. Note that the center of gravity is the combined center of gravity for: the bucket, the attachment bracket and the load. The equations for the position of the center of gravity are:

ˆ

x : Lx= L1cos(θlif t) + L2cos(α + ϕtilt) (4.31)

ˆ

z : Lz= L1sin(θlif t) + L2sin(α + ϕtilt) (4.32)

where α is the angle from bottom of the bucket and up to the center of gravity.

L1 is the length of the boom and L2is the distance from the end of the boom to

the center of gravity. The position of the center of gravity is not estimated after each bucket fill but is seen as a fix point that has been calculated with a fully and evenly distributed load in the bucket. The center of gravity for only the bucket and load where found Latin.

(33)

4.5 Center of gravity 21

Figure 4.5: Figure from which (4.31) and (4.31) can be derived. Note that the tilt angle (ϕ) is zero in the figure.

(34)

(35)

Chapter 5

How To Empty The Bucket

To be able to answer the question: How to empty the bucket? one must first know what mandatory requirements there are and what is considered good performance when loading the hauler in the short loading cycle. Three interviews and one preliminary study have been made to create a framework and to find the best solution for the problem. The three people that where asked to participate are all Volvo CE employees with different backgrounds, all with a lot of experience of wheel loaders. These people are: Gunnar Löwestand, Joakim Unnebäck and Torbjörn Martinsson. A summary of the interviews is presented in Section 5.1 below. The preliminary study is a comparison of two methods that after the interviews seemed interesting, these are presented in Section 5.3.

5.1 Interviews

The summary of the interviews is divided into two sections. The first section is called Framework where the questions concerning setup, where to put the different buckets on the hauler et cetera are brought up. The second section is execution and that is the section that will define how to perform the bucket emptying in the short loading cycle.

5.1.1 Framework

This section defines the setup and where to put the load on the hauler. It also explains the matching, safety and other key topics that need to be taken into account when loading a hauler in the short loading cycle.

The setup

The setup in shown in Figure 5.1 where the angle β ∈ (45◦, 60◦). The back end of the hauler shall be close to the pile to minimize the distance traveled by the wheel loader. The setup is one of the key factors for a fast and energy efficient loading cycle. It is therefore the driver of the wheel loader that decides where the

(36)

24 How To Empty The Bucket

hauler shall be placed. How the pile looks like is another key factor but this is not a part of this thesis, but how e.g. angle of the pile and mean particle size effect productivity can be studied in [7].

Figure 5.1: The setup used in the short loading cycle.

Load the hauler

Where to place the buckets on the hauler differs between drivers, but they all put some of the early buckets in the back of the hauler to prevent the last bucket to overflow at the back. It is not uncommon for drivers to spill some. However it is important to keep the ground as flat as possible, too much material spillage and time has to be wasted on leveling the ground. Gunnar Löwstrand’s way of loading the hauler is shown in Figure 5.2. To get the material centered on the hauler and to prevent spillage, the bucket shall be parallel to the edge of the hauler when material starts to fall out. The hauler itself is not that sensitive to oblique loading but no material shall fall out during transportation.

Figure 5.2: The figure shows how Gunnar Löwstrand fills the hauler. He puts the first bucket at the back (to the right in the figure) of the hauler.

(37)

5.1 Interviews 25

Matching

There are some things that need to be matched when loading a hauler in the short loading cycle.

Matching - bucket size, wheel loader size and hauler size. Is about matching the bucket depending on material to the wheel loader and to match the wheel loader to the hauler. Since the wheel loader always fill the bucket to approximate 110% of the buckets volume and the wheel loader can only carry that much weight the size of the bucket needs to match the material and the wheel loader. The L120 can e.g. lift about 6000 kg (by law when using a bucket) and depending on the density of the material the size of the bucket need to match so that a full bucket weighs 6000 kg. With gravel a normal size for the bucket is 3.6 cubic me-ters. There is also matching, for the same reason as above, with hauler size and wheel loader size. The hauler shall be filled with an integer-number of buckets, meaning an A25 (can carry 25000 kg) needs four buckets from an L120 to be filled. Matching - driving speed and lifting speed. The driving speed needs to match the lifting speed to prevent unnecessary braking at the end. If a driver faces the problem that the bucket will not achieve the desired height in time it is normal to step on the brake while keeping the same engine speed and sometimes even a higher engine speed is used. The brakes has to be used because if the driver only uses the throttle to get a higher engine speed (and lifting speed) the driving speed will increase more than the lifting speed since the lifting speed is a linear function of the engine speed while the driving speed is nonlinear (approximately a second order polynomial). In a worst case scenario the driver can apply the brakes so hard that the wheel loader will stop completely while keeping the same engine speed to ensure the lifting speed is still high.

Safety

The highest risk is that of tipping over, either to the side or forward. The L120 has a tipping load of 12000 kg and with that load and the bucket straight out (θlif t = 0◦) from the machine, it will tip forward. Already with a 6000 kg load

and heavy braking there is a risk of tipping over. The machine is also quite unstable with the bucket high above the ground and especially with a large waist angle. Uneven ground can of course also be a problem with the stability. To read more about the stability of wheel loaders see [8].

Productivity versus efficiency

The efficiency must not lower the productivity! In this thesis the productivity can basically be viewed as an engine speed, since that will set a time limit for the driving towards the hauler. So given an engine speed the way of emptying the bucket shall take as little time as possible to achieve a high productivity. The hauler is very expensive when it is standing still. Normally the productivity is measured as ton per hour and efficiency as litre per ton or ton per litre.

(38)

Abrasion

In the short loading cycle there is not much to think about except the obvious, like too high engine speed will cause unnecessary braking, abrasion on the converter and engine et cetera.

5.1.2 Execution

The sections here are divided into different states, compare with Algorithm 4.1.

Reversing

Since the reversing point is depending on the driving distance it is important to have the bucket at a height that matches the distance, thus to be able to drive short distances the bucket needs to be as high as possible in the reversing point.

Translate towards the hauler/receiver

Adjust speed, with the brake or by changing the engine speed, to match the lifting of the bucket. Even though some driver use braking in this state, it is not recommended to use the brakes as a way to match the driving speed and lifting speed. Braking (to stop in front of the hauler) should be done as late as possible. It might also be considered a good idea to use the fact that the wheel loader in this state has a large roll resistance. The positioning of the machine relative the hauler should be that the edge of the bucket is parallel to the edge of the hauler and the waist angle should be as straight as possible for increased stability.

Emptying the bucket

The most important task here is to get the load centered on the hauler without damaging the hauler. With some material e.g. blasted rocks, the first bucket must be laid down slowly to create a protective layer. The last bucket might need some more attention to prevent spillage. There will always be some material that falls on the ground the important thing is that this material does not build up and makes the ground bumpy. The angle at which the material starts to fall out differs a lot between different materials.

Translate from the hauler/receiver

First tilt back the bucket and reverse the wheel loader back to the reversing point. Note that this reversing point is not necessary the same as when exiting the pile. Maximum lowering speed of the bucket is not dependent of the engine speed, which means that the matching problem does not exist in this state.

(39)

5.2 Demands on the system 27

5.2 Demands on the system

From the interviews a few extra demands on the system came to light, which are presented below as parameters that needs to be set before each emptying. These parameters are:

• Tilt angle at the hauler’s edge, • the height of the end position,

• parameter to reduce emptying speed for first bucket, • the distance on which the emptying shall be done, • deceleration distance.

5.3 Preliminary study

From the interviews two different methods of emptying the bucket seemed to be interesting to compare. One method (Section 5.3.1) that only uses one function at a time and another method that allows for simultaneous usage of lift and tilt functions (Section 5.3.2). In this preliminary study only the hydraulic system were studied, which means that the matching problem temporarily is overlooked. The idea for the path planning (to the hauler) is however to take the shortest path, which means that the hydraulic system will be the limiting factor. In the prelim-inary study the engine speed will be constant (to get an idea of how long time it will take to raise the bucket) and equal in both methods. The path driven is assumed to be straight (no hydraulic flow to the steering).

Both methods have the same initial and final position but have different basic ideas on how to move the bucket to the right position. The initial position has been taken from a measurement of an experienced driver and the final position is the position where the wheel loader stands when the material begins to fall out of the bucket and onto the hauler. These angles are

• Start position: θlif t= −13◦ and ϕtilt= 64◦

• Final position: θlif t= 20◦ and ϕtilt= 0◦

To be able to compare these methods to each other, similar Simulink models have been used. Both models used the same geometry blocks (in Simulink) and both used flow calculations to ”move” the bucket. The difference between the methods lies in how the flow is directed.

5.3.1 Method 1 - Separately

The first method is the simple one and inexperienced driver usually use this way because it is easier to control the machine and it gives a larger margin to the hauler. The basic idea here is to use the lift and tilt functions separately, that

(40)

is, first use the lift until the bucket is clear of the hauler’s edge. Then the driver simply tilts the bucket when the tip of the bucket is over the hauler’s edge. Some drivers lowers the bucket again just before the material starts to fall out of the bucket to ensure the material is placed right on the hauler. The lowering of the bucket before the tilting start is not a part of method. This method is similar to the one that Volvo CE/Swecon promotes in their basic driving instruction video on the website [1]. This method has been chosen because of mainly two reasons.

• It is simple (easy way of driving).

• It only uses one function at the time (no losses in the hydraulic system due to simultaneously lifting and tilting).

Deeper analysis of this method revealed some drawbacks. The biggest draw-back lies in how the center of gravity moves when the lifting has finished and the tilting starts. The Figure 5.3 shows that the center of gravity is lifted unneces-sarily high and it also shows the path of the center of gravity during the tilting. The extra work that this leads to with a load of 5500 kg can theoretical be cal-culated with Eloss,1= mg∆h = 54.1 kJ where m is the mass of load, bucket and

attachment bracket (7732 kg).

ï20

0

20

40

60

80

3.2

3.4

3.6

3.8

4

4.2 Movement of centerïofïgravity during tilt

[m]

[degree]

Figure 5.3: The path of the center of gravity during tilting with lift angle θ = 20◦. Starts at the plus sign in the upper right corner and stops at the triangle in the bottom left corner. The difference in height is ∆h = 0.71 m.

(41)

5.3 Preliminary study 29

5.3.2 Method 2 - Simultaneously

In the second method lifting and tilting will be done simultaneously even though there will be losses in the hydraulic system due to the pressure difference in the tilt and lift cylinder as described in Section 4.4. Since the problem with method one was that of that the center of gravity was lowered during the tilting, method two will use lift and tilt to keep the center of gravity at a constant height while rotating the bucket around that point, see Figure 5.4.

Figure 5.4: Method two - rotate the bucket around the center of gravity. So in the second method the lift is used to, first lift the center of gravity to the height that the center of gravity will have when the material starts to fall out of the bucket. Then secondly when the center of gravity has reached that height the lift and tilt will be used to keep the center of gravity constant, or as it is implemented, keep the velocity of the center of gravity to zero in the z-direction. To be able to estimate the pressure in the cylinders an approximation that the bucket will move with constant velocity was made. Then the pressures at different lift and tilt angles could be found in Latin (at Volvo), the hydraulic flow where calculated in the Simulink model. The results from these calculations can be found in Figures 5.5, 5.6 and 5.7 and the energy losses is according to (4.24) equal to

Eloss,2= 31.5 kJ. 0 1 2 3 4 0 10 20 30 40 50 60 [s] [degree]

Lift and tilt angles

\_tilt e_lift

(42)

Figure 5.5: Shows how the lift and tilt angles is changing during the simultaneously lifting and tilting from simulations in Simulink.

0 1 2 3 4 0 0.5 1 1.5 2 2.5 3 [s] [ dm 3/s ]

Flow in lift and tilt cylinders q_T q_L

Figure 5.6: Hydraulic flow in the lift and tilt cylinder during the simultaneously lifting and tilting from simulations in Simulink.

0 0.5 1 1.5 2 2.5 3 13.4 13.6 13.8 14 14.2

Pressure in lift cylinder

[MPa]

[s]

0 0.5 1 1.5 2 2.5 3

5 10

Pressure in tilt cylinder

[MPa]

[s]

Figure 5.7: The pressure in the lift and tilt cylinder during simultaneously lifting and tilting from simulations in Simulink and data from Latin.

5.3.3 Conclusions from the preliminary study

From the preliminary study some advantages and disadvantages have come up with both methods. The major advantages of method one has already been mentioned and these are:

• it is easy,

• hydraulic system efficient.

If all the available flow is used to lift, tilt or lift and tilt, the both methods will take just as long to finish. The reason for that is because the piston in the cylinder

(43)

5.4 Solution - How to empty the bucket 31

has to move exactly the same length, which corresponds to a volume of hydraulic oil that needs to be pumped into the cylinder. Now if the flow is the same in both methods they have to take equal time to finish. However some material needs to be laid down with care and that is the same as to make sure the material gets as little momentum as possible. With these materials method one that has a higher center of gravity when the tilting starts and uses all the available flow to tilt needs to reduce the flow to the tilt much more then method two needs to. Method two also uses all the flow but has divided that flow to both lift and tilt meaning this method is faster with these materials. To summarize the biggest disadvantages with method one:

• the center of gravity is raised too high,

• more time consuming (depending on material).

In the second method the biggest advantage is that it seems to be more en-ergy efficient, by comparing the enen-ergy losses from method one with method two:

Eloss,1 = 54.1 kJ and Eloss,2 = 31.5 kJ. However since there were some

approx-imations done in the preliminary study this result is not as accurate as it could be said that method one has an energy loss that is 22.6 kJ larger. But the fact that the energy loss in method one is almost twice that of method two is definitely something that shows that method two is more energy efficient than method one. Then there is something that perhaps should not matter but does so in this case since the autonomous wheel loader is being developed as a demonstrator and that is the visual appearance. Method two looks more professional and is therefore better. The second method’s advantages can be summarized into:

• more energy efficient,

• faster (depending on material), • visually better.

The drawbacks of this method have mostly been mentioned before and are: • complex,

• losses in the hydraulic system.

A lot points towards method two but how complex method two will be when implementing it should not be underestimated. There is a big difference between controlling one function at a time and controlling both simultaneously in the way that the control system is designed (due to coupling between the lift and tilt).

5.4 Solution - How to empty the bucket

The conclusion from the interviews and the preliminary study shows that method two is most suitable for this application even though it is harder to implement. It was decided that the system has to be able to calculate a starting condition for

(44)

when the wheel loader can begin to move towards the hauler without unnecessary braking. That means that if the hauler is too close when the wheel loader at the reversing point, the wheel loader will adjust the lift angle in the reversing point before starting to move towards the hauler. Then, when the starting condition is met the wheel loader can start to move forward. This method of standing still and lifting is not optimal but reversing even more is of course worse. An experienced driver should in this case have lifted the bucket higher going out from the pile. The autonomous machine can not do this when the thesis is written because the bucket will then be in the way of the laser scanner (the wheel loader must scan the hauler to be able to calculate a path to it), however this will hopefully soon be solved and the autonomous machine will also be able to lift the bucket while going out from the pile.

The demands presented in Section 5.2 are of course part of the solution. The hydraulic flow has been chosen to be the key variable to control and the machine will always use all the available hydraulic flow at all time because the pumps are the most efficient when their full displacement are used. In the future, changing the engine speed can perhaps solve the matching problem.

(45)

Chapter 6

The Design

In this chapter the bucket emptying function and its implementation will be ex-plained.

6.1 Setup

The hauler is placed with an angle of about 50◦ and about 1.5 meters from the pile (see Figure 5.1. This setup has been chosen to be able to set parameters in the autonomous machine that has nothing to do with this thesis. The important issues concerning the setup in this thesis is that the path to the hauler is not too long, because then the setup will be more like the load and carry cycle, which is not a part of this thesis.

6.2 Bucket path planning function

The bucket path planning function takes the planned way to the hauler, engine speed and model parameters as input. The models parameters mainly consist of the parameters mentioned in Section 5.2. The planned path (now on called waypoints) consist of the distance from the reversing point and the current waist angle, the engine speed is also set for each waypoint. The model parameters are parameters to change the max lift height, the hauler’s height, reduce the available flow, reduce the flow during the emptying part, the tilt angle at hauler’s edge, parameter that can shift the flow between lift and tilt during emptying, the emp-tying distance and the deceleration distance.

To plan the bucket’s path three Simulink models have been used, powertrain simulation, lift model and emptying model. The bucket path planning function will return a pair of angles (lift and tilt) for each waypoint given as input and the angular velocity for lift and tilt plus the angles that the wheel loader shall have when the specified height is reached. Note that the bucket path planning function only returns values that corresponds to a waypoint, which will have the

(46)

34 The Design

effect that the wheel loader can reach the final waypoint (at the hauler) but still has not reached the final position with the bucket. In that case the wheel loader automatically creates ramps from the current positions to the final positions that it will carry out while standing still.

It was also decided that 10% of the available flow should be reserved for the control system. If the bucket path planning function would have used 100% of the available flow the control system has no flow to use if one of the functions (due to model error) falls behind and need extra flow, which means that it will be very sensitive.

6.2.1 Powertrain simulation

To estimate how much hydraulic flow that is available in (or more accurately, in between) each waypoint, it is essential to estimate the time difference between each waypoint. This is done by simulating the powertrain with a Simulink model. The powertrain is modeled as described in Section 4.3 and the roll resistance parameters in (4.7) can be used to tune the powertrain simulation. The input to this model is the waypoints and the engine speed (assumed to be constant), while the output is the time difference between waypoints and the wheel loaders velocity. The powertrain simulation part of the Simulink model can be seen in Figure 6.1 that is named vehicle_model.

[rad/s^2] [rad/s] wheelspeed to turbinespeed Distance 1 vehicle_model T_t ang_vel Distance ang_acc rpm_to_n rpm_to_n i Product Memory DiscreteïTime Integrator K Ts zï1 Converter n_pump n_turbine T_turbine Compare To Zero < 0 final_dist 2 RPM 1

(47)

6.2 Bucket path planning function 35

This model only simulates with a positive acceleration. The braking to stop before the hauler is calculated with a Matlab function, which is then added to the list of time differences from the Simulink model.

6.2.2 Lift model

One of the demands on the system was that the bucket emptying function should return a starting condition if the planned engine speed and way to the hauler is enough to lift the bucket as intended. It was also decided that the wheel loader should remain in the reversing point while lifting the bucket to the angle set as starting condition. Due to this fact the lift model has a reversed causality, mean-ing the initial position for this model is the angles that the bucket will have at the hauler’s edge. The model will then calculate backwards using all the available hydraulic flow to the reversing point and if the simulated lift angle and the actual lift angle are equal, then the starting condition is met. However, if the simulated lift angle is higher then the actual lift angle the simulated angle will be the start-ing condition and the lift has to reach that angle in the reversstart-ing point before it can start to move towards the hauler. When the distance to the hauler is long the simulated lift angle will be reached at some point between the reversing point and the hauler. In that case the wheel loader will keep the lift angle it had in the reversing point as long as possible.

The available hydraulic flow that is mentioned above is the total available flow from the pumps according to (4.19) minus the flow to the steering. The flow to the steering is estimated with a model of the steering cylinders and the fact that the waist angle and therefore the waist angular velocity is known, which can be translated into hydraulic flow using (4.29) modified for the steering cylinders.

The lifting of the bucket to the edge of the hauler is divided into two steps. 1. Lifting the center of gravity to the right height.

2. Use lift and tilt to rotate the bucket around the center of gravity.

During the first step all the available flow is used in the lift cylinder to lift the center of gravity to the height that it will have at the hauler’s edge. In the second step the center of gravity is kept at a constant height until the bucket has reached the hauler’s edge (where material starts to fall out) and the emptying model takes over. To keep the center of gravity at a constant height, the derivative of (4.32) together with the (4.29), (4.30) and (4.20) forms the equation below that must hold:

0 = L1sin(θlif t)

qmax− qtilt

glif t(θlif t)Acyl,lif t

+L2sin(α+ϕtilt)

qtilt

gtilt(θlif t, ϕtilt)Acyl,tilt

(6.1)

6.2.3 Emptying model

When emptying the bucket on the hauler both lift and tilt is used. In this model there is also extra degrees of freedom to be able to get the material centered and

(48)

36 The Design

to be able to put the loads carefully on the hauler. The extra degrees of freedom are:

• Reduce available flow.

• Decide the percentage of the available flow that’s going to the tilt (and lift). • The lift angle at the end (the buckets maximum height).

The model divides the total available flow (sometimes reduced) between the lift and tilt cylinders.

6.3 Implementation

To be able to implement the design on the autonomous wheel loader the Simulink models needed to be compiled to stand-alones and then be included in the C# project that also includes the GUI. This had to be done in three steps.

First all three Simulink models had to be compiled to stand-alone files using RSIM in Simulink. RSIM allows Simulink models to read input from a file and all the parameters in the models can be changed without having to recompile the models. Secondly a few Matlab functions ware created, three functions to run each now a stand-alone Simulink model (set parameters and input) and one master Matlab function to run the other three Matlab functions and which will return the output. Thirdly the master Matlab function was compiled using Matlab Builder NE to a dll-file, which then were included in the C# project.

6.4 Control system for lifting and tilting

The original control system was kept but some problems occurred during the im-plementation and the root of these problems was mainly due to the extra flow to the tilt cylinder during simultaneous lift and tilt. The problem comes from model error in the open loop control due to that the pressure difference over the tilt valve is higher then normal. During simultaneous usage of these functions the pressure difference in the cylinders will make the flow to the tilt cylinder to be more than intended and thus the lift will move slower or stop completely.

Another problem was that the control system did not take into account the effect lifting has on the tilt angle. During lifting (and lowering) of the bucket the tilt will have a positive (negative) angular velocity. This made the control signal from the open loop control inaccurate (when lifting and tilting simultaneously). Since there are significant time delays in the system it is important that the open loop control is accurate to make the closed loop function properly.

(49)

6.4 Control system for lifting and tilting 37

6.4.1 Innovations

The existing control system had to be modified to suit the needs of the new way of emptying the bucket. The major innovations to the control system are listed below.

Open loop controller

The problem that extra flow goes to the tilt cylinder is dependent on the pressure difference over the tilt vale, which is not known during simultaneously lifting and tilting. Therefore this problem has to be handled by the control system using feedback from the angular sensors. As an extra precaution, a way of prioritizing the lift has also implemented.

Prioritized lift

The lift is prioritized if the lift cannot follow its path during simultaneously lifting and tilting. To make more flow go to the lift a PI-controller will use the error from the lift to reduce the control signal from the tilt regulator. The tilt will then slow down and more flow is available to the lift. This is only used as a precaution and because of that, there is a lower limit on the absolute value of the error from the lift that needs to be exceeded before the PI-controller is activated.

Lifting’s affect on the tilt angle

During lifting the tilt angle will change and it is easy to realize if the tilt is unused during lifting (see Figure 1.1). However the control system did not take that into account, which made the path following quite bad. To remove the influence the lifting had on the tilt angle a geometric model was created that calculated the ”extra” tilt angle, which then were deducted from the original tilt angle.

Control direction

During bucket emptying if one of the functions overshoots the control system should not correct the error by changing the sign of the control signal, e.g. by lowering the bucket or tilt back the bucket, since that would be very inefficient with respect to energy usage. To solve this there are predefined direction in which the control system can work. So now when the lift or tilt overshoots the target it will keep that value until it is back on the path.

(50)

(51)

Chapter 7

Results

The results are divided into three parts 1. powertrain simulation,

2. path following, 3. the center of gravity.

The engine speeds (1200, 1400 and 1600 rpm) have been chosen because lower speeds than 1200 rpm are too slow and higher then 1600 rpm is outside the eco-nomic range. The measurements made at 1200 rpm were measured at Tech-show in Gothenburg, where the autonomous wheel loader was shown for selected Volvo AB employees, customers and media. This set of data has been tuned to be as robust as possible, while the data from 1400 and 1600 rpm prioritizes productivity and efficiency. All the data is collected when performing an actual short loading cycle, which means that the load in the bucket, the traveled distance et cetera vary for each set of data.

7.1 Powertrain simulation

The simulations and measurements from the powertrain simulations are shown in Figures 7.1-7.3. The traveled distance is shown together with the simulated distance as a function of time. In Figures 7.1 and 7.2 the simulated distance is at some points greater than the actual. The effect that this will have is that the wheel loader will have more time between waypoint and therefore will have no problem to reach the right angles. Vice versa if the simulated distance is less than the actual the control system has to compensate for that error but the powertrain model is tuned so that this case is very rare.

The reason that the simulated distance in 1200 rpm is that much greater than the measured is because it was tuned for the Tech Show and then robustness was of the highest priority. In Figure 7.3 it is clear that the final distances vary between

(52)

40 Results

the simulated and the measured, which is probably (at least partially) due to that the wheel loader failed to stop at the right place.

0 2 4 6 8 10 0 1 2 3 4 5 6 7 8 9 Powertrain simulation, 1200 rpm Time [s] Distance [m] Measured Simulated

Figure 7.1: Results from the powertrain simulation with 1200 rpm.

0 2 4 6 8 0 2 4 6 8 10 Powertrain simulation, 1400 rpm Time [s] Distance [m] Measured Simulated

(53)

7.2 Path following - angles 41 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8 Powertrain simulation, 1600 rpm Time [s] Distance [m] Measured Simulated

Figure 7.3: Results from the powertrain simulation with 1600 rpm.

7.2 Path following - angles

In Figures 7.4 - 7.6 the calculated paths (setpoint) and the measured angles are shown for both lift and tilt. The measured angle corresponds well with the calcu-lated path for all three speeds.

A deeper analysis of the figures reveals that the part of the path that originates from the emptying model is almost completely absent. The reason for that is not because it does not perform well itself, but rather the fact that the brakes sometimes are not able to keep the intended speed during the last braking. This made the emptying somewhat unstable since the actual speed did not match the simulated. The brakes are very complex and hard to control so instead of tuning the brakes it was decided that during the last braking the wheel loader should follow a ramp instead. Ramps are independent of the actual speed and therefore made the emptying smoother, but it became harder to get the load centered on the hauler.

(54)

42 Results 0 2 4 6 8 10 12 ï50 0 50 Lift angle, 1200 rpm Time [s] Angle [degree] 0 2 4 6 8 10 12 ï100 0 100 Tilt angle Time [s] A ng le [d egree ] Setpoint Measured Error

Figure 7.4: Results from the lift and tilt angles in 1200 rpm. The error is the difference between the setpoint and the measured value.

0 2 4 6 8 ï50 0 50 Lift angle 1400 rpm Time [s] Angle [degree] 0 2 4 6 8 ï50 0 50 Tilt angle Time [s] Angle [degree] Setpoint Measured Error

(55)

7.3 Center of gravity 43 0 1 2 3 4 5 6 ï50 0 50 Lift angle, 1600 rpm Time [s] Angle [degree] 0 1 2 3 4 5 6 ï50 0 50 Tilt angle Time [s] Angle [degree] Setpoints Measured Error

7.3 Center of gravity

The Figures 7.7 - 7.9 show the part of the path where the center of gravity should remain constant. The angles from this part of the path have been put into (4.32) to get the height over the ground. The path in Figure 7.8 is somewhat more unstable then the others, which can be explained by that the hydraulic oil in the wheel loader was at the time of measurement not warm enough.

0 0.5 1 1.5 2 2.5 3.6 3.8 4 4.2 4.4

Center of gravity (measurements), 1200 rpm

Time [s] [m] 0 0.5 1 1.5 2 2.5 3.6 3.8 4 4.2 4.4

Center of gravity (setpoint) , 1200 rpm

Time [s]

[m]

(56)

44 Results 0 0.5 1 1.5 2 3.6 3.8 4 4.2 4.4

Time [s] [m] 0 0.5 1 1.5 2 3.6 3.8 4 4.2 4.4

Center of gravity (setpoint), 1400 rpm

Time [s]

[m]

Figure 7.8: The path of the center of gravity at 1400 rpm.

0 0.5 1 1.5 3.6 3.8 4 4.2 4.4

Time [s] [m] 0 0.5 1 1.5 3.6 3.8 4 4.2 4.4

Center of gravity (setpoint), 1600 rpm

Time [s]

[m]

(57)

Chapter 8

Conclusion and Future Work

The purpose of this thesis was to find out how to empty the bucket onto a hauler in an efficient way with high productivity and then implement that solution on the autonomous wheel loader that is developed by Volvo CE in Eskilstuna.

8.1 Conclusion

It has been proven in this thesis that there are both time and energy to save depending on how the bucket is emptied on the hauler. From the two methods that are compared it is clear that the method of keeping the center of gravity at a constant height is more efficient and sometimes (depending on material) faster than the method that only uses the lift and the tilt separately.

The method of keeping the center of gravity was implemented on the au-tonomous wheel loader and was used to fill a hauler in front of an audience at Tech Show in Gothenburg. After Tech Show there where only a little time left for validation before the wheel loader had to be disassembled for renovation. This led to that we only validated the system while loading three buckets (and not four) on the hauler, which we did with great success. To validate the fourth bucket no modifications should be necessary in the bucket path planning function. The problem most likely will be to plan where on the hauler to put the four loads so that the hauler will not overflow.

8.2 Future work

To improve the bucket emptying further, here are some suggestions:

• Model the powertrain so that it can use varying engine speeds as input. • System for detecting spillage.

• Safety system that use the brakes if the bucket will hit the hauler. 45

(58)

46 Conclusion and Future Work

(59)

Bibliography

[1] Grundläggande körtekniker hjullastare. http://www.volvoce.com. (Swedish). [2] Jonathan Blom. Autonomous hauler loading. Master’s thesis, Mälardalen

University.

[3] Volvo CE. Volvo wheel loaders - L110F, L120F. search for L120F on http://www.volvoce.com/.

[4] Lars Eriksson and Lars Nielsen. Modeling and control of engines and drivlines. 2009.

[5] Reno Filla. Operator and machine models for dynamic simulation of construc-tion machinery. PhD thesis, Linköpings Universitet, 2005. ISBN 91-854557-14-0.

[6] Noriho Koyachi and Shigeru Sarata. Unmanned loading operation by au-tonomous wheel loader. ICROS-SICE International Joint Conference, 2009. [7] S.P. Singh and R Narendrula. Factors affecting the productivity of wheel

loaders in surface mines. International Journal of Mining, Reclamation and Environment, 20, 2006. ISSN 1748-0930 print or ISSN 1748-0949 online. [8] Toyota. C-truckar - Utbildning. Toyota - Material handling. ISBN:

970-91-978655-2-4, (Swedish).

[9] Takashi Tsubouchi, Shigeru Sarata, Hisashi Osumi, and Masamitsu Kurisu. Introduction of yamazumi project - trial for autonomous heavy vehicles at construction site -. Industrial Technology, 1, 2002.

(60)

Autonomous bucket emptying on hauler

Institutionen för systemteknik

Department of Electrical Engineering

Examensarbete

Autonomous bucket emptying on hauler

Autonomous bucket emptying on hauler

Examensarbete utfört i Reglerteknik

vid Tekniska högskolan i Linköping

av

Abstract

Acknowledgments

Contents

Chapter 1

Introduction

1.1

Problem formulation

1.2

Limitations

1.3

Thesis outline

1.4

Definitions

Chapter 2

The Machine - L120F

2.1

Power distribution

2.2

Hydraulic system

2.2.1

Priority in the work hydraulics

2.3

Engine

Chapter 3

The Autonomous Wheel

Loader Project

3.1

Previous work

3.1.1

Additional sensor

3.1.2

Control system

3.2

Brakes

3.3

The goals with the autonomous wheel loader

3.4

Related work

Chapter 4

Background Theory

4.1

Short loading cycle

4.2

Coordinate system

4.3

Automotive modeling

4.3.1

Torque converter

4.3.2

Transmission and final gear

4.3.3

Wheel

4.3.4

Complete model

4.4

Hydraulics

4.4.1

Hydraulic losses

4.4.2

Lift and tilt cylinder

4.4.3

Steering cylinders

4.5

Center of gravity

Chapter 5

How To Empty The Bucket

5.1

Interviews

5.1.1

Framework

5.1.2