Dynamic Torque Control

of Brushed DC Motors
for Hardware-in-the-Loop Integration

Eric Schyllert

Department of Automatic Control

MSc Thesis
TFRT-6220
ISSN 0280-5316

Department of Automatic Control

Lund University
Box 118
SE-221 00 LUND
Sweden

© 2023 Eric Schyllert. All rights reserved.

Printed in Sweden by Tryckeriet i E-huset
Lund 2023
Abstract

The main goal with the thesis was to design, construct and evaluate a system able
to drive and brake a mechanical axis to simulate loads and stored rotational energy
- a hardware force simulator. The force simulator was then to be integrated in a
existing hardware-in-the-loop testing rig to enhance the mechanical dynamics of
door simulations. The force simulator was to be realised using a brushed DC motor
and by controlling the motor torque. Controlling the motor torque of a brushed DC
motor is achieved by controlling the motor current - and this is mainly what this
thesis is about. The thesis explains the foundations of an embedded system capable
to control and log the current of a brushed DC motor.
The force simulator consisted of a microcontroller, a motor card, a circuit to
measure the inline motor current and a brushed DC motor. The circuit to measure the
current was designed and implemented, consisting of a shunt resistor and a current
sense amplifier. The shunt was placed in series (inline) with the motor. A software
strategy was developed and implemented to deal with noise due to common-mode
voltage transients (caused by motor control using PWM). This strategy came with
a cost of introducing a measurement delay in the system.
A discrete PI-controller to control the motor current was researched and imple-
mented. An expression to optimal tune the controller based on motor parameters and
the sampling period was researched. Current control experiments were conducted
but due to a calculation error the PI-controller was very untuned, this compared to
the tuned expression. This error was discovered very late in the work and there was
no time to redo the experiments. The experiments with the implemented untuned PI-
controller were conducted and a current reference was successfully tracked, how-
ever with noise which had an amplitude of approximately 200 mA.
A strategy to record current measurements used in the current control, without
missing any samples was successfully developed. It worked by using a double buffer
system, filling the buffers utilizing a DMA and then written to a SD card once a
buffer was filled. With the main objective to develop the capability to be able to
analyze the current control algorithm.

3
Acknowledgements

I would like to express my heartfelt gratitude to my company supervisors Eric Warn-

quist and Per Byrhult for their guidance and support throughout my project. They
made the daily work so much more fun. I would also like to extend my apprecia-
tion to Roland Larsson - Electronics Engineer at ASSA ABLOY, for the guidance
when I faced challenges with electronics. But most importantly, for citing the quote
"Assumption is the mother of all fuck ups" - a quote that was proven true numer-
ous times throughout the course of my thesis work. I would also like to express my
gratitude to all the people at ASSA ABLOY who have contributed to creating an
exceptional place to work at.
To my supervisor from LTH, Anton Cervin - thank you for always being sup-
portive. I would also like to thank Karl-Erik Årzén for being the examiner of this
thesis.
An immense gratitude is expressed to my family for their unwavering support,
especially to my brother who solely steered my academic path.

5
Contents

1. Introduction 9
1.1 Problem formulation . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2. Background 12
2.1 Measuring current using a current shunt . . . . . . . . . . . . . 12
2.2 Torque control - brushed DC motor . . . . . . . . . . . . . . . . 12
2.3 Sampled current control - brushed DC motor . . . . . . . . . . . 13
2.4 Simplified model of the mechanics of a brushed DC motor . . . . 15
3. System overview 16
3.1 Motor setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Motor driver card . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Microcontroller . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Micro SD card . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.5 Operating the system . . . . . . . . . . . . . . . . . . . . . . . 18
4. Current Control 21
4.1 Current shunt placement . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Analog-to-digital convert with the Arduino MKR Zero . . . . . . 22
4.3 Current sense amplifier . . . . . . . . . . . . . . . . . . . . . . 23
4.4 Current sense dead-zones and PWM dependency . . . . . . . . . 25
4.5 Current control overview . . . . . . . . . . . . . . . . . . . . . 27
4.6 Challenges in current control of unloaded DC motors . . . . . . 27
5. Motor Control 29
5.1 Motor control strategy . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Encoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6. Software design 33
6.1 Software overview . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.2 Direct memory access controller . . . . . . . . . . . . . . . . . 33
6.3 PWM generation . . . . . . . . . . . . . . . . . . . . . . . . . . 33

7
Contents

6.4 Strategy to go from the case "slow computer" to "fast computer" 35

6.5 Current average calculations . . . . . . . . . . . . . . . . . . . . 38
6.6 Strategy to write to SD card in real-time . . . . . . . . . . . . . 38
7. Results 42
7.1 Final iteration of current measurement setup . . . . . . . . . . . 42
7.2 Dead-zone strategies . . . . . . . . . . . . . . . . . . . . . . . . 43
7.3 SD card benchmark evaluation: assessing write speed and latency 44
7.4 Writing to SD card in real-time using DMA . . . . . . . . . . . 45
7.5 Current average calculations . . . . . . . . . . . . . . . . . . . . 45
7.6 Strategy to go from the case "slow computer" to "fast computer" 45
7.7 Current control with blocked rotor . . . . . . . . . . . . . . . . 46
8. Discussion 50
8.1 Final iteration of current measurement setup . . . . . . . . . . . 50
8.2 Dead-zone strategies . . . . . . . . . . . . . . . . . . . . . . . . 50
8.3 Current averaging calculations . . . . . . . . . . . . . . . . . . 51
8.4 SD card benchmark evaluation: assessing write speed and latency 51
8.5 Writing to SD card in real-time using DMA . . . . . . . . . . . 51
8.6 Strategy to go from the case "slow computer" to "fast computer" 52
8.7 Current control with blocked rotor . . . . . . . . . . . . . . . . 52
9. Conclusion 54
9.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
9.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
A. Code 56
A.1 Tuned control algorithm . . . . . . . . . . . . . . . . . . . . . . 56
A.2 Simplified and untuned control algorithm . . . . . . . . . . . . . 57
Bibliography 58

8
1
Introduction

ASSA ABLOY Entrance Systems has a wide range of automatic door systems,
each product with its own unique mechanical dynamics. Today at Assa Abloy their
physical testing rig does not take into account these mechanical difference between
different door systems. The motor in the testing rig is not loaded with any physical
load. The testing rigs are small and fairly mobile and enables developers/testers to
do verification and testing of new implementations without having to access a real
door system.
A more realistic testing rig would make it easier to motivate the use of it in a
third party certification process of the door system. One such step in the certification
process is to ensure that the control system and motors are operable in temperatures
of -30 °C to 50 °C. Placing a fully installed door system in a climate chamber is a
hard task - if the testing rig was used instead it would make it a lot easier and speed
up the process, reduce the cost and open up the use of smaller climate chambers.
A more accurate testing rig in which a range of different door types and config-
urations can be hardware simulated opens up for a more dynamic and agile verifi-
cation and testing process in which the control system can be tested for edge cases
such as really heavy doors or a door system installed with a floor material with a
really high frictional coefficient. This could speed up development time due to de-
creasing time from implementation to verification, make testing more accurate and
increase the trust in the testing process from a third party certification point of view.
The main goal with the thesis was to design, construct and evaluate a prototype
system able to drive and brake a mechanical axis to simulate loads and rotational
energy stored in the system. This load simulator would be used to simulate the
dynamics of doors and to be integrated in a existing hardware-in-the-loop testing
rig - to capture a more realistic behaviour.
The scope was narrowed due to time constraint to create a system able to mea-
sure motor current and to log the data in real-time, and to control the current of a
brushed dc motor. Due to the physics of a brushed dc motor, to control the motor
current is equivalent with being able to control the motor torque which would be
used to drive a mechanical axis to simulate stored rotational energy. The logging of
data would be used to be able to validate the current control.

9
Chapter 1. Introduction

1.1 Problem formulation

The unloaded door system is simplified as to be viewed as a speed controlled motor
- defined as the system motor. The unloaded system motor has dynamics that differ
from when it is loaded with the physical door. The real door system can have up
to two motors working together but in this thesis it is simplified as only having
one. The door system is using CAN (Controller Area Network) to receive the speed
reference.
To hardware simulate the stored rotational energy a force driving the mechanical
axis connected to the system motor is needed. To achieve this a brushed DC motor is
used. An initial rough idea of how a control system to simulate the rotational energy
in the door system can be designed can be seen in Figure 1.1 in which five key
components can be identified: mathematical model/recorded current measurements,
CAN bus, system motor, physical mechanical axis and a simulator motor together
with a control system.

.
Figure 1.1 Overview of what a control system to simulate rotational energy as-
sumed could be composed of. Five key components were identified: mathematical
model/recorded current measurements, CAN bus, system motor together with the
control system under test, physical mechanical axis and a simulator motor with a
control system.

The idea was that the error between the mathematical model/recorded values
and the actual values of the system motor was to be used as a reference to the
simulator motor which would be controlled to minimize the error between the cur-
rent/speed of the system motor and the model. It was limited to only compensate for

10
 1.2 Methodology

the error when it meant that the simulator motor would apply a driving force that
would increase the speed of the system motor - simulating stored rotational energy.

1.2 Methodology
The work methodology during the thesis work was applied practical work with an
iterative approach with each iterative phase consisting of research, prototyping and
verification. Both hardware and software verification was often done with an oscil-
loscope by capturing signals and evaluating the result. A signal generator, a power
supply unit and a multi-meter was also used in the verification process.

1.3 Thesis outline

The first part of the thesis commences with an introduction that presents the
project’s background, objectives and a rough idea of how this can be achieved.
Chapter 2 presents background theory that a lot of the work is based on. Chap-
ter 3 presents an overview of the system with each subsections detailing each key
component that the system is composed of. Chapter 4 presents how current control
of a brushed DC motor will be achieved and defines goals based on the current con-
trol. Chapter 5 discusses motor control, how to operate the motor card and relates
how a current reference results in a duty cycle. Chapter 6 details how the software
is designed. Chapter 7 presents the results from different experiments and calcula-
tions. Chapter 8 discusses the results presented in chapter 7. Chapter 9 contains the
conclusions made from the work and research done. It also presents the limitations
and what is left to do.

11
2
Background

2.1 Measuring current using a current shunt

To measure current using a shunt resistor the key principle is that the current flows
through the shunt resistor creating a small voltage drop across the resistor that is
then measured. The size of the current shunt is limited by the maximum power
dissipation that it is rated to handle [Itarsiwala, n.d.] and how big intrusiveness that
the shunt is allowed to have on the voltage across the motor due to the voltage
drop caused by the shunt. Equation 2.1 shows how to calculate the maximum power
dissipation based on resistance size and the maximum current that will flow through
the shunt.
2
Power_Dissipation(W )max = Rshunt Iload_max (2.1)

2.2 Torque control - brushed DC motor

To be able to control the torque generated by the brushed DC motor a model of the
torque is required. Equation 2.2 represents a simplified model of a brushed DC mo-
tor - a model that explains the dynamics of the motor current [Alaküla and Karlsson,
2011].

di
Udc = RI + L − ωψ (2.2)
dt

where Udc represents the DC input voltage, R is the electrical resistance of the mo-
tor’s armature, I is the electrical current flowing through the armature, L is the in-
di
ductance of the motor’s armature, dt is the rate of change of current with respect to
time, ω is the angular velocity of the motor’s armature and ψ is the magnetic flux
linkage in the motor’s armature.

12
 2.3 Sampled current control - brushed DC motor

Equation 2.3 expresses the relationship between the motor current and the motor
torque [Alaküla and Karlsson, 2011] - if the goal is to control the motor torque it
requires to be able to control the motor current.

τ = Iψ (2.3)

where τ represents the generated torque, I is the electrical current flowing through
the motor’s armature, and ψ is the magnetic flux linkage in the motor’s armature.

2.3 Sampled current control - brushed DC motor

When calculating the current reference for discrete current control the concept of
fast and slow computer arise. The difference is a matter of when the sampled current
is used to calculate the current reference. The definition of a fast computer is when
the computer is able to provide a current reference for a sampling interval, that is
based on measurements sampled in the beginning of the sampling interval. On the
other hand, a slow computer provides a current reference based on the measurement
of current taken at the beginning of the previous sampling interval [Alaküla and
Karlsson, 2011].
For sampled current control with a ’fast computer’ a proportional integral feed-
forward (PIF) controller is derived in [Alaküla and Karlsson, 2011]. The controller
is derived based on the simplified model of the DC-motor shown in Equation 2.2
and under the assumptions listed below:

ū(k, k + 1) = u∗ (k) (2.4)

∗
i(k + 1) = i (k) (2.5)
i∗ (k) + i(k)
ī(k, k + 1) = (2.6)
2
ē(k, k + 1) = e(k) (2.7)
n=k−1 
i(k) = ∑ i∗ (n) − i(n) (2.8)
n=0

where k indicates the beginning of the sample interval, ū(k, k + 1) is the average
voltage during sample k, u∗ (k) is the reference voltage, i(k) is the measured current,
i∗ (k) is the reference current, ī(k, k + 1) is the average current during sample k,
ē(k, k + 1) is the average back-emf during sample k and e(k) is the measured back-
emf. Assumption 2.4 means that the reference voltage for the coming period should
be the average voltage needed to achieve the desired current change. Assumption 2.5
is an assumption of "dead-beat" current control - meaning that the current error is
eliminated in one sampling interval. Assumption 2.6 is argued to be true in average.
Assumption 2.7 is usually true due to that the sampling frequency of the current
control is magnitudes faster than the dynamics of the back-emf. Assumption 2.8

13
Chapter 2. Background

is a consequence of the assumption of "dead-beat" current control [Alaküla and

Karlsson, 2011]. The PIF-controller is tuned based upon the motor parameters and
the sampling time of the current control. The expression for the tuned PIF-controller
is shown in Equation 2.9:

!
L n=k−1
∗ R  ∗  Ts ∗

u (k) = + · i (k) − i(k) + L Ts · ∑ i (n) − i(n) + e(k)
Ts 2 | {z } |R + 2 n=0 {z |{z}
} Feed forward
Proportional
Integral
(2.9)
where u∗ (k) represents the voltage reference, R is the electrical resistance of the
motor’s armature, L is the inductance of the motor’s armature, Ts is the sampling
period, i∗ (k) is the current reference, i(k)is the measured
 current, e(k) is the back-
emf added as a feed-forward and ∑n=k−1
n=0 i∗ (n) − i(n) represents the accumulated
current errors.
Equation 2.10 shows the basic equation for a discrete PI controller derived using
Euler transformation [Bengtsson, 2020].
 Ts n=k 
u(k) = K p e(k) + ∑ e(n) (2.10)
TI n=0

where u(k) is the discrete control signal, e(k) is the error between the reference
and the actual value of the sampled process, ∑n=k n=0 e(n) is the accumulated errors,
K p is the amplification, TI is the integration time and Ts is the sample time.
By comparing Equation 2.9 with Equation 2.10 and setting e(k) = i∗ (k) − i(k),
e(n) = i∗ (n) − i(n) we see that, if we disregard the feed-forward in Equation 2.9,
that these two equations are almost the same. The boundary of the summation differs
with one sample. We identify that K p = TLs + R2 and TI = L 1 Ts .
R+ 2

Figure 2.1 Basic control system - inspired by [Bengtsson, 2020]

14
 2.4 Simplified model of the mechanics of a brushed DC motor

2.4 Simplified model of the mechanics of a brushed DC

motor

If the rotor of a brushed DC motor is simplified as single axis lumped inertia drive
then the motion is described by Equation 2.11 [Leonhard, 2012]

dω
J = mM (ω,t) − mL (ω,t) = ma (ω,t) (2.11)
dt
where J represents the moment of inertia of the rotor, ω represents the angular
velocity of the rotor, t represents time, mM (ω,t) represents the torque due to the
motor current, mL (ω,t) represents the torque due to the load and ma (ω,t) represents
the resulting torque on the system. Such a simplified single axis lumped inertia can
be seen in Figure 2.2.

Figure 2.2 This figure illustrates a simplified model of the mechanics of the rotor
when simplified as a single axis lumped inertia drive - inspired by [Leonhard, 2012]

15
3
System overview

The system marked with dashed lines in Figure 1.1 is the system that will be used
to simulate stored rotational energy. This system consist of a microcontroller, a mo-
tor setup with encoder, an evaluation module (DRV8704EVM) based on the motor
driver DRV8704, a micro SD card, a current measuring circuit and a circuit to sup-
ply pull-up resistors for the signals from the motor encoder. This is the system that
was worked on in this thesis and an overview can be seen in Figure 3.1.

Figure 3.1 System overview of a system capable of current control in a dc motor

and real-time

16
 3.1 Motor setup

To simplify the process of the laboratory work a command prompt was made
which enabled the user to send input through the serial port to the microcontroller
to carry out commands and to display reply messages. All time critical operations
are interrupt driven which means that the command prompt is not able to steal valu-
able processor time from important operations. To get a better understanding of the
system a list with information about all interrupts can be seen in Table 6.1. Writing
data to the SD card can be seen as a time critical operation - it is not interrupt driven
- and can thus get CPU time hijacked from interrupts. The interrupts are designed
to be executed as fast as possible and the importance of being quick increases with
increasing calling frequency.

3.1 Motor setup

An initial existing setup with two brushed DC motors connected on same axis was
used as the basis of the thesis work which can be seen in Figure 3.2.

Figure 3.2 Existing setup with two brushed DC motors connected on same axis.
Mechanical axis marked in black and motors color-coded - red defined as simulator
motor and blue as system motor.

3.2 Motor driver card

A customer evaluation module designed around the motor driver DRV8704 was
purchased and used in the project. The DRV8704 is a dual H-bridge brushed DC
motor driver, it is highly configurable and supports running two different brushed
DC motors.

17
Chapter 3. System overview

Figure 3.3 Figure showing the DRV8704EVM - a platform built to be able to drive
two different brushed DC motors with variable current limiting using a dual H-bridge

3.3 Microcontroller
The Arduino MKR Zero was selected and used in this project to act as the micro-
controller with the purpose to orchestrate and carry out necessary calculations and
operations. The processor of the Arduino MKR Zero is the Arm Cortex-M0 32-bit
SAMD21 processor. The processor contains peripherals with key functions such as
timers, signal generators and analog-to-digital converters that were utilized during
the prototype building. An overview of the microcontroller and how it is connected
can be seen in Figure 3.4.

3.4 Micro SD card

Due to the Arduino MKR Zero having a limited amount of memory - 256KB Flash,
32KB SRAM [SAM D21 Family Data Sheet 2018] an external 32 GB flash memory
was purchased and used. This to be able to log current measurements sessions of
varying lengths, to later be able to verify the results of the current control algorithm.

3.5 Operating the system

To operate the system the command prompt is used, common usage could be: turn
on/off the motor, calibrate no current voltage, set/get a current reference, start a
session to log data to the SD card, display errors from the DRV8704EVM or to
configure the DRV8704EVM - a list of all commands can be seen in Table 3.1.

18
 3.5 Operating the system

Figure 3.4 Overview of Arduino pins and what they are connected to. *Only one
of the PWM signals goes to D5 due to lack of OR circuit.

19
Chapter 3. System overview

Command Arg 1 Arg 2 Return Value Side Effects Description

motor on/off void changes regis- Turn the motor on
ter CTRL or off
readReg [0-7] int [0-65535] Read selected
register from
DRV8704
writeReg [0-7] [0- int [0-65535] Write to se-
4096] lected register of
DRV8704
current read on void Writes to a Starts storing the
predefined file voltage used to
on the SD card, calculated the duty
reads a global cycle to a SD card
variable
current get int Reads a global Retrieves the cur-
variable rent reference
current set x void Writes to a Sets the cur-
[x<=0] global variable rent reference,
currently only
implemented to set
negative values
current calibrate int Changes regis- Turns off the
ter CTRL motor, waits until
it is stopped and
then measures the
no current voltage,
updates a global
variable with the
new value

Table 3.1 Table with information about all commands implemented in the system
to use in the command line used to operate the system

20
4
Current Control

To be able to control the motor current it is necessary to be able to measure the actual
motor current. A system to measure the motor current - with sufficient amplitude
resolution and sample frequency was necessary to be designed and implemented.
The current control in the real door system served as a requirements baseline - with
an current amplitude resolution of roughly estimated 400 mA and a PWM frequency
of 25 KHz - in which the duty cycle is updated at a frequency of 5 KHz.
The main goal was to be able to measure the inline motor current with precision
of around 400 mA, bidirectional currents of maximum 16.5 A in both direction and
be designed to be able to handle higher currents up to 32 A without breaking. The
ambition was to create a prototype fast that could be evaluated and later improved
upon. The current measurement had to be able to measure bidirectional currents due
to the real door system being able to rotate in both directions.
The shunt size was chosen to be able to handle twice the maximum current of the
maximum current of the motor in the real door system with Equation 2.1 in mind.
The goal was to be able to use the same measuring circuit for the real system and
a motor operated at twice the maximum current (twice as strong) without breaking.
The motivation was to prototype faster and not having to design more than one
measurement circuit.

4.1 Current shunt placement

The DC motor can be seen as an inductive load and will resist a change in the
current flowing through it and it is common practice to place the motor in a H-
bridge. There are three potential locations that the current shunt can be placed to
measure the current - low side (in series with ground), high side (in series with the
voltage supply) or inline (inside the H-bridge in series with the motor). The only
shunt placement that measures the current going through the motor at all times is
the inline placement with which the true phase current can be known which gives us
the ability to monitor the system. The inline implementation however comes with

21
Chapter 4. Current Control

the problem that if using a current sense amplifier it has to support high common-
mode rejection [Bridgmon and Andrews, 2016].

4.2 Analog-to-digital convert with the Arduino MKR Zero

The Arduino MKR Zero has an ADC peripheral that was configured to sample
periodically at 250 K samples/s with 12 bit resolution and in differential mode -
calling an interrupt every time a measurement is made. The Arduino’s 3.3 V output
was used as the voltage reference. The requirement on the ADC was to be able to
catch the dynamics of the motor current such that a closed loop current control could
be realized. The initial goal was to have a closed loop current control with an update
frequency of 5 KHz. The Nyquist’s theorem state that the sample frequency has to
be twice that of the signal to be recreated [Glad and Ljung, 2003] - meaning a lower
limit of sampling frequency of 10 KHz. The possibility to average measurements to
reduce noise was also a motivation to keep the sampling frequency high.

It was observed that the voltage measured with the Arduino ADC had a constant
gain error, this was verified by measuring the voltage with a voltmeter and then
comparing it with the voltage measured with the Arduino - for different voltages.
The real voltage was that of 87.5% of what the Arduino measured, Equation 4.1
compensate for this gain error:

Voltagereal = 0.875 · ADCvoltage (4.1)

For an n bit ADC the number of discrete digital levels that it can produce is given
by:

ADClevels (n bits) = 2n (4.2)

The resolution of the smallest voltage that can be quantified is then given by:

Re f erence Voltage Re f erence Voltage

ADCresolution (n bits) = = (4.3)
ADClevels (n bits) 2n

22
 4.3 Current sense amplifier

4.3 Current sense amplifier

Due to worldwide semiconductor chip shortage it was very hard to find components
and some compromises in design had to be made. One such compromise was that
the Current Sense Amplifier INA240 was only available in the INA240A3DR ver-
sion - which has a gain of 100. The high gain meant that a small shunt had to be used
to not saturate the output of the current sense amplifier. A shunt of one milli Ohm
was purchased and deemed "good enough" - due to the time consuming process of
finding components that were in stock. The Current Sense Amplifier INA240A3DR
has variable supply voltage and can be operated from 2.7 V to 5.5 V - this was one
of the parameters in consideration when selecting Current Sense Amplifier with the
goal to operate it at 3.3 V. The Arduino MKR Zero used in the project has a 3.3 V
logic level and with a risk of damage if operated at higher voltages. Not operating
the Current Sense Amplifier at a higher voltage level than 3.3 V reduce the risk
of an accident in which the output from the Amplifier destroys the Arduino. The
maximum voltage range of the ADC in the Arduino MKR Zero is 3.3 V, the current
Sense Amplifier operated at 3.3 V is then making full use of the entire range of the
ADC.

The motor current is symmetrical and bidirectional which means that:

Imax = −Imin (4.4)

The ADC is only able to measure non-negative voltages and to be able to measure
negative voltages the Current Sense Amplifier has to be biased such that one part of
the voltage range is used to represent negative currents and the other part positive
currents. The Current Sense Amplifier was biased at half the supply voltage due to
the currents being symmetrical. The region of no output saturation is then described
by:

Vsupply
0≤ + Ishunt Rshunt · gain100 ≤ Vsupply (4.5)
2
From 4.5 the maximum current without output saturation is described by:

Vsupply
Imax ≤ (4.6)
2Rshunt gain100

Calculating the maximum current according to Equation 4.6 when Vsupply =

3.3 V, Rshunt = 1 mΩ one obtains:

3.3 V
Imax = = 16.5 A =⇒ Imin = −16.5 A (4.7)
2 · 1 · 10−3 Ω · 100 Equation 4.4

23
Chapter 4. Current Control

Using Equation 4.3 for a 12 bit ADC and a reference voltage of 3.3 V, the voltage
level resolution for the ADC is calculated as:

3.3 V
= 0.806 mV (4.8)
212

The smallest input voltage to the Current Sense Amplifier that the ADC is then
able to quantify and thus also the smallest current this corresponds to when using
Rshunt = 1 mΩ is given by:

0.806 mV 8.06 µV
= 8.06 µV =⇒ I = A = 8.06 mA (4.9)
gain100 1mΩ
This means that the smallest current the system theoretically can measure with-
out regarding any errors is 8.06 mA.

24
 4.4 Current sense dead-zones and PWM dependency

4.4 Current sense dead-zones and PWM dependency

The current sense amplifier - INA240 can sense drops across shunt resistors over a
wide common-mode voltage range from -4 V to 80 V [INA240 2016]. The common-
mode voltage range is important due to the current shunt being inline with the motor
and due to this will be exposed to the large DC voltage used to supply the motor.
The enhanced PWM rejection provides high levels of suppression for large
common-mode transients ( ∆V ∆t ) [INA240 2016] - these are caused by the switch-
ing and occur when the PWM-signal changes state that changes the state of the
H-bridge.
The common-mode transients due to a common-mode voltage step can be seen
in Figure 4.1 and Figure 4.2 and there is a zone in which the current sense amplifier
output is completely unusable due to having a very large error - this area could be
regarded as a measuring dead-zone. The common-mode steps occur when the PWM
signal changes state: from low to high or high to low.

Figure 4.1 Print-screen from oscilloscope: Green signal shows the PWM-signal,
olive green signal shows the output from the current sense amplifier. The area inside
the red cursors enclose the time span with very large output error from the current
sense amplifier caused by a change in the state (high to low) of the PWM-signal.

25
Chapter 4. Current Control

Figure 4.2 Print-screen from oscilloscope: Green signal shows the PWM-signal,
olive green signal shows the output from the current sense amplifier. The area inside
the red cursors enclose the time span with very large output error from the current
sense amplifier caused by a change in the state (low to high) of the PWM-signal.

To deal with the "dead-zones" a strategy to discard measurements was imple-

mented. There can be up to two dead-zones during every PWM-period. One dead-
zone is stationary and can only occur at the start of the PWM-period. The other
dead-zone is time varying and dependent of the duty cycle - it can occur at any time
during the PWM-period. To deal with the stationary dead-zone three samples are al-
ways skipped at the start of the PWM-period - independent if needed or not. It was
identified that the time varying dead-zone only could occur in the transition from
high to low PWM-signal. To deal with time varying dead-zone a interrupt attached
to an analog input of the Arduino was implemented. The interrupt tells the ADC to
skip three samples, it is triggered by falling edge of the analog input and the analog
input is fed the PWM-signal - which is when the varying dead-zone is active.

26
 4.5 Current control overview

4.5 Current control overview

If we could achieve current control with "fast computer" a current control model was
proposed and can be seen in Figure 4.3 and it is inspired by Equation 2.9 and Figure
2.1. The controller is a discrete PI-controller which outputs a voltage reference.
The back-emf is calculated and added as a feed-forward to the voltage reference.
The sample time of the loop was assumed to be in the range of 4-6 kHz while the
calculated back-emf would be updated with a frequency of 100-200 Hz. The duty
cycle algorithm is to control the H-bridge in such a way that the supply voltage in
average is the voltage reference across the motor during one PWM period.

Figure 4.3 Overview of the closed current control loop in the case of "fast com-
puter"

4.6 Challenges in current control of unloaded DC motors

Using Equation 2.11 as a model for the rotor and inserting the force produced by
the motor we get Equation 4.10 which describes the acceleration of the rotor.

dω τmotor − τload
= (4.10)
dt J
where τmotor is the force produced by the DC-motor, τload is the force counter-
acting the motor - such as friction, J the moment of inertia of the rotor and dω
dt is the
rate of change of the motor speed.

27
Chapter 4. Current Control

Rewriting Equation 2.2 to get the expression of the rate of change of the motor
current one obtains:

di Udc − RI − ωψ
= (4.11)
dt L
When a DC motor is unloaded, there is very little mechanical load on the motor
shaft, and the motor can spin at a very high speed with very little current flow-
ing through the electrical resistance of the motor’s armature. This can be seen by
combining Equation 4.11 and Equation 4.10 - if a small current accelerate the rotor
speed to a high speed the resulting back-emf will be large and oppose the flow of
current through the motor’s armature. The back-emf will limit how big the current
can grow.

28
5
Motor Control

To supply the voltage and current required by the motors, a motor driver circuit is
necessary. This circuit also serves to separate and protect the microcontroller from
the high currents and voltages used with DC motors. The DRV8704 used in the
project can handle a supply voltage of 8 V to 52 V. The chosen motor card can
be configured via SPI, making it easy to modify its behavior without needing to
replace hardware. Another highly beneficial feature of the motor driver card is its
capability to report errors. This information can be accessed through the SPI inter-
face, enabling easy detection and resolution of issues without the need for hardware
replacement.

5.1 Motor control strategy

Figure 5.1 Table of how to operate the motor using the motor driver DRV8704
and two PWM-signals xIN1 and xIN2. The red zones marks the modes used in the
project.

29
Chapter 5. Motor Control

To operate the motor, two PWM signals (xIN1 and xIN2) and two GPIO signals
(X and Y) were utilized. The motor output was controlled with the PWM-signals
according to Table shown in Figure 5.1. The GPIO signals X and Y were used to
control each of the half-bridges which can be seen in Figure 5.2

Figure 5.2 Overview of signals controlling the H-bridge - PWM-signals: xIN1,

xIN2 and GPIO-signals: X, Y

An algorithm of how to control the PWM-signals and GPIO-signals based on

the voltage reference u∗ (k) was constructed. The algorithm is based upon activating
forward current (xIN1 and Y used), reverse current (xIN2 and X used) or H-bridge
disabled. The algorithm can be seen in Equations 5.1, 5.2, 5.3 and 5.4.

30
 5.1 Motor control strategy

 u∗ (k)
∗
 VV M , 0 ≤ u (k) ≤ VV M

D1 (u∗ (k)) = 1, u∗ (k) > VV M (5.1)

0, u∗ (k) < 0

 u∗ (k)
∗
− VV M , 0 > u (k) ≥ −VV M

D2 (u∗ (k)) = 1, u∗ (k) < −VV M (5.2)

0, u∗ (k) ≥ 0

(
∗ 1, u∗ (k) ≥ 0
X(u (k)) = (5.3)
0, u∗ (k) < 0
(
0, u∗ (k) ≥ 0
Y (u∗ (k)) = (5.4)
1, u∗ (k) < 0

where VV M is the DC supply voltage, D1 (u∗ (k)) controls the duty cycle of PWM-
signal xIN1 and D2 (u∗ (k)) controls the duty cycle of PWM-signal xIN2.

31
Chapter 5. Motor Control

5.2 Encoder
An incremental encoder attached on the motor rotor was used - producing two
square-wave signals phased at 90 degrees which are defined as the alpha and beta
signals and that can be used to determine the rotational direction of the motor and
motor position. Each motor revolution produces 500 alpha and beta impulses.
The encoder is using a logical level of 5 V which is to high too be used directly to
the Arduino MKR zero which uses 3.3 V logic. The encoder is using an open-drain
output that can only sink current, meaning we need an external source to supply
current when the signal is high. Figure 5.3 shows the final iteration of the encoder
circuit with a voltage dividers to transform the 5 V logic down to approximately 3.3
V and with pull-up resistors to supply current when the signal is high.

Figure 5.3 Shows the encoder circuit with voltage dividers and pull-up resistors -
with the encoder output marked as Alpha and Beta

32
6
Software design

6.1 Software overview

The software is designed to run on a single core microcontroller and were everything
time critical is interrupt driven. The software is designed to control a control system
that controls the simulator motor. The control system can be seen in the red box in
Figure 1.1. Setting the correct priority of interrupts is crucial in order to ensure that
the most time-critical actions are performed with little to no delay. Table 6.1 shows
the interrupts used in the control system, their priority and if they were periodic or
event driven.

6.2 Direct memory access controller

The direct memory access controller (DMAC) contains both a direct memory access
(DMA) engine and a cyclic redundancy check (CRC) engine. The DMAC is used
to move data between memory and peripherals and is able to do this without the
involvement of the CPU [SAM D21 Family Data Sheet 2018].
The motivation behind using the DMAC was mainly to offload the CPU but also
to use it as a tool to synchronize the flow of certain operations - by having events
generate interrupts, events such as a buffer being filled. This was later utilized to
write to a SD-card in real time.

6.3 PWM generation

Two instances of the Timer/Counter for Control applications (TCC) peripherals in
the microcontroller were configured to perform waveform generation - to create
three PWM signals. By configuring the counters with the same internal clock and
prescaler and configuring the counters to count clock pulses results in the two coun-
ters being synchronized. The output of the wave generated is controlled by a register
(CCx) that the current count value is compared with, if the current count value is

33
Chapter 6. Software design

Interrupt Name Interrupt Description Triggered by

Priority
TCC0_Handler() 0 Calculates and sets the duty Periodic every 41.67 µs
cycle every fourth PWM
period.
TCC1_Handler() 1 Tells the ADC to ignore x Periodic every 41.67 µs
amount of measurements, a
strategy to deal with the sta-
tionary dead zone
ADC_Handler() 0 Called every time the ADC Periodic every 4 µs
makes a measurement
TC4_Handler() 3 Calculates the motor speed Periodic every 200 ms
based on the change of
alpha counts between each
TC4_Handler() interrupt
call
skipSamples() Same as the Tells the ADC to ignore x Event: triggered by
number of amount of measurements - falling PWM edge max
the pin used strategy to deal with vary- every 41.67 µs
ing dead zone
alphaInterrupt() 3 Used to count alpha pulses Event: triggered by ris-
from encoder ing edge from encoder -
max every 35.82 µs
DMAC_Handler() 0 Used to set flags indicating Event: called when
that a buffer is filled DMAC buffered filled

Table 6.1 Showing information about all interrupts implemented in the system

higher than the register CCx the output is low otherwise the output is high. The
counter will count up to the value set by the register PER and then restart the count
from zero.

Equation 6.1 express the relationship between the counters clock frequency,
prescaler and the maximum count value (PER) and the frequency of the generated
PWM signal [SAM D21 Family Data Sheet 2018]. The counters were configured to
use a 48 MHz clock frequency, prescaler=1 and maximum count value PER=2000
which using Equation 6.1 results in a PWM frequency of 24 KHz.

fclock 48MHz
fPW M = = = 24KHz (6.1)
Prescaler · PER 1 · 2000

34
 6.4 Strategy to go from the case "slow computer" to "fast computer"

Equation 6.2 expresses the relationship between the compare value CCx and the
maximum count value PER that the counter is configured to use and the duty cycle
of the generated PWM signal.

CCx
Duty cycle = (6.2)
PER
One of the counters (TCC1) was configured to generate two of the three PWM
signals - these two signals share the same count value but their output are respec-
tively controlled by different compare registers (CC0 and CC1). The two PWM
signals are defined as xIN1, xIN2 and are fed to the motor driver - thus the signals
controlling the DC motor. The signals can be seen in overview Figure 3.1 and in
Figure 5.1 in which they show how their state relate to the motor output. As the
count value reaches the maximum count value PER an interrupt is called - in the
case of the counter TCC1, TCC1_handler() is called, more information about the
interrupt can be seen in Table 6.1.

6.4 Strategy to go from the case "slow computer" to

"fast computer"
To motivate the use of the ’fast computer’ model the assumption was that the duty
cycle calculations had to be done close in time to their actual usage. To achieve
this a second counter called TCC0 was introduced. The TCC0 was fed the same
clock as counter TCC1 - thus synchronized but placed out of phase to be triggered a
time before the TCC1 counter. The reason was to have time to calculate and set the
duty cycle of the next PWM period of TCC1 before TCC1 starts outputting the next
PWM period, the process is shown in Figure 6.1. An other important reason to do
this is to be able to use current measurements taken closer in time to the next PWM
period. If the interrupt TCC1_Handler() would have been used to calculate and set
the duty cycle - then the current measurements would have been at least one PWM
period delayed from when the new duty cycle could have been used.

By updating the duty cycle every fourth PWM period the resulting sample period of
the duty cycle is described by Equation 6.3.

Ts = 4 · TPW M (6.3)

Using the proposed strategy with synchronized counters to time the calculation of
the new duty cycle - the time between starting to calculate the new duty cycle and
when the new duty cycle is put in use is described by Equation 6.4.

OFFSET
dt = · TPW M , where OFFSET ≤ PER (6.4)
PER

35
Chapter 6. Software design

Figure 6.1 Strategy using synchronized counters to time the calculation of the next
duty cycle closer in time to its usage

where dt is the time to calculate the new duty cycle, OFFSET is the added
count value to offset the second counter TCC0, PER is the maximum count value
and TPW M is the period of the PWM-signal.

How the sample period of the duty cycle relates to the calculation time and the
period of the PWM signal can be seen in Figure 6.2. By comparing the time to
calculate the duty cycle expressed in Equation 6.4 in relation to the sample time of
the duty cycle which is expressed in Equation 6.3 - we get the relative duty cycle
calculation delay which is described in Equation 6.5

OFFSET
dt PER · TPW M OFFSET
= = , where OFFSET ≤ PER (6.5)
Ts 4 · TPW M 4 · PER

where dt
Ts is the relative calculation delay. A relative calculation delay of one
would be the case "slow computer" discussed in the background chapter and a rela-
tive delay of approximately zero would be the case of "fast computer". This under
the assumption that the current measurement used in the duty cycle calculation is
taken in the beginning of the calculation.

36
 6.4 Strategy to go from the case "slow computer" to "fast computer"

Figure 6.2 Overview of how the sample period Ts of the duty cycle is related to the
period of the PWM signal TPW M . The figure shows the time interval dt that represents
the delay from when the duty cycle is calculated until it is put in use.

37
Chapter 6. Software design

6.5 Current average calculations

Values from the ADC were continuously updating a global voltage average. A bit-
shift averaging algorithm was used to calculate the average voltage and can be seen
in Equation 6.6. A fast algorithm was assumed to be needed due to having voltage
measurement taken periodically every 4 µs.
(
acc = (voltage_avg(k) << n) − voltage_avg(k))
(6.6)
voltage_avg(k + 1) = (acc + measurement) >> n

acc denotes the accumulator variable, measurement denotes the newest input sam-
ple from the ADC and n is a factor determining how much the new measurement
will impact the average. This algorithm is commonly used to perform digital filter-
ing of data samples in real-time applications, where the goal is to smooth out the
signal while minimizing processing overhead. The resulting filter introduces a delay
in the system.

6.6 Strategy to write to SD card in real-time

The goal was to store the voltage used in every duty cycle calculation without miss-
ing any values. The duty cycle is updated with a frequency of 6 KHz but the actual
interrupt is called with a frequency of 24 KHz and in every fourth interrupt the duty
cycle is calculated.

The interrupt is called every 6 KHz and the data to be stored is 4 bytes in size, the
generated data rate is then calculated as:

KB
6·4 = 24 KB/s (6.7)
S

The rate of the data generated in Equation 6.7 is a slower than the write speed to the
SD card seen in Table 7.1, which is required to not miss generated data. Designing
for the worse case scenario the maximum writing latency from Table 7.1 is 12.368
ms but this is in a benchmark environment and not tested in the load simulator
environment - in which CPU time is taken by interrupts. Assuming a maximum
latency of 250 ms, a minimum array size required to be able to store all samples if
each sample is taken with a frequency of 6 KHz is calculated as:

variables
Bu f f ermin = Latencymax · = 250 ms · 6 KHz = 1500 (6.8)
time

38
 6.6 Strategy to write to SD card in real-time

A double buffer system was designed and implemented to be able to save data with
the DMA while writing to the SD card simultaneously. The buffer size was chosen
to 2560 with Calculation 6.8 in mind. The time to fill the buffer with the DMA
saving one value with a frequency of 6 KHz is calculated as:

2560
= 426.66 ms (6.9)
6KHz

During this time the SD card with a latency of 250 ms and write speed of 480 KB/s
can write an amount of bytes calculated as:

(426.66 − 250) ms · 480 KB/s = 84796.8 ≈ 84796 bytes (6.10)

The amount of bytes that an array of datatype Long and size 2560 takes up is cal-
culated as:

2560 · 4 byte = 10240 bytes (6.11)

To structure the data such that it can be understood when exported from the SD card
extra symbols are required to be written, if an extra 4 bytes is added with every data
point then the total data to be written to the SD card from an array of size 2560 is
calculated as:

2560 · (4 + 4) bytes = 20480 bytes (6.12)

The time to fill one buffer of 2560 samples when samples are stored with a
frequency of 6 KHz is calculated according to Equation 6.9 to 426.66 ms. In theory,
the double buffer system should work because the amount of data required to empty
a buffer of 2560 samples, including extra characters, is determined to be 20480
bytes according to Equation 6.12. This quantity is smaller than the data that can be
written to the SD card in 426.66 ms according to Equation 6.10. Therefore, it is
feasible to implement the double buffer system, which allows for continuous data
filling in one buffer while the other buffer is being written to the SD card.
From the command prompt the user starts the data logging, this activates the
DMA, the DMA is configured such that it is triggered by the interrupt that calcu-
lates the duty cycle - when triggered it saves the voltage used for the duty cycle
calculation. A flowchart of the cyclic process of the DMA filling two buffers can
be seen in Figure 6.3. The DMA interrupt is called when a buffer is filled, a flag is
set indicating that it is ready to be written to the SD card. The interrupt checks if
the SD card is done writing to the next buffer - if it is not done an error is incre-
mented, indicating that samples might have been missed due to the DMA filling and
SD writing is overlapping - the flowchart of this erroneous behaviour can be seen in
Figure 6.5 and Figure 6.4 shows the desired behaviour.

39
Chapter 6. Software design

Figure 6.3 Flowchart of DMA configured in cyclic mode and set to suspend the
DMA and trigger a DMA interrupt when a buffer is filled - the interrupt executes
some operations and then reactivates the DMA and the process of filling the next
buffer is started.

Figure 6.4 Time chart of how the DMA and writing to SD card operates in time
when working as expected.

40
 6.6 Strategy to write to SD card in real-time

Figure 6.5 Time chart of how the DMA and writing to SD card operates when a
buffer takes to long to write to the SD card and an error is generated. The SD card is
writing from the buffer that the DMA has started to fill with new data - this result in
the risk overwriting data not yet recorded by the SD card.

41
7
Results

7.1 Final iteration of current measurement setup

The final iteration of the current measurement setup consists of a current shunt
placed inline with the motor - connected via a Kelvin clutch. The voltage across the
shunt is then fed to a bidirectional current sense amplifier together with the ADC of
the Arduino configured in differential input. The a schematic of the system can be
seen in Figure 7.1.

Figure 7.1 Schematic of the current measuring circuit - ADC+ and ADC- goes to
the ADC differential inputs of the Arduino.

42
 7.2 Dead-zone strategies

7.2 Dead-zone strategies

To evaluate the effectiveness of the dead-zone strategies three different experiments
were conducted. Both the varying and stationary dead-zone strategy were config-
ured to skip two samples. At the time of the experiment the strategy to write to a
SD card in real-time using DMA had not been implemented. Due to this the mea-
surements were taken once the motor had reached a stationary speed for a given
constant duty cycle. Each experiment contained of two different measuring inter-
vals, each measuring interval captured continuous samples. The first interval was
current measurements when the duty cycle was set to 0%, this to show the noise not
caused by a switching PWM-signal. The second measuring interval was with a duty
cycle of 75% this to show the noise caused by a switching PWM-signal.
The first experiment was conducted with no algorithm to skip samples, the sec-
ond experiment with the algorithm to deal with the stationary dead-zone and the
third experiment with the algorithm to skip both the varying and stationary dead-
zone. In Figure 7.2 measurements were made without any method to disregard dead-
zone samples, in Figure 7.3 the method to disregard samples from the stationary
dead-zone was implemented and in Figure 7.4 both the method to disregard dead-
zone samples from the stationary and the varying dead-zone were implemented.

Figure 7.2 Print Screen from the Arduino IDE, no implemented method to deal
with dead-zones. Note: the y-axis is the current measured in mA, the X-axis is dis-
continuous and can be broken up in two continuous parts one part (x values from
1554 to 1620) showing continuous samples with duty set to zero and the other part
(x values from 1621 to 2054) continuous samples with duty set to 75%. The blue
graph shows the measured current.

43
Chapter 7. Results

Figure 7.3 Print Screen from the Arduino IDE, Stationary dead-zone skip method
implemented. Note: the y-axis is the current measured in mA, the X-axis is discon-
tinuous and can be broken up in two continuous parts one part (x values from 778 to
910) showing continuous samples with duty set to zero and the other part (x values
from 911 to 1270) continuous samples with duty set to 75%. The blue graph shows
the measured current.

Figure 7.4 Print Screen from the Arduino IDE, Stationary and varying dead-zone
skip methods implemented. Note: the y-axis is the current measured in mA, the X-
axis is discontinuous and can be broken up in two continuous parts one part (x values
from 716 to 920) showing continuous samples with duty set to zero and the other part
(x values from 921 to 1216) continuous samples with duty set to 75%. The blue graph
shows the measured current.

7.3 SD card benchmark evaluation: assessing write

speed and latency
An open-source library [Greiman, 2021] was used to test the write speed and writing
latency of the micro SD card and the result from the tests can be seen in Table 7.1.
The test was done when the microcontroller ran a benchmark software and not the

44
 7.4 Writing to SD card in real-time using DMA

load simulator software.

Write Speed Max Latency Min Latency Average Latency

Iteration 1: 488.14 KB/s 4875 us 1044 us 1046 us
Iteration 2: 487.57 KB/s 12368 us 1044 us 1047 us
Table 7.1 SdFat micro SD card benchmark results

7.4 Writing to SD card in real-time using DMA

Using two buffers of size 2560 the strategy to write to SD-card in real time worked
without generating any errors. This while logging measurements from the voltage
used to calculate the duty cycle - which was recorded with a frequency of 6 KHz.

7.5 Current average calculations

A moving averaging filter was used to smooth out the motor current measurements.
The filter used is presented in Equation 6.6 and the speed of the filter used was set
to n=1. The resulting filter is derived in Equation 7.1:
(
acc = (voltage_avg(k) << 1) − voltage_avg(k))
(7.1)
voltage_avg(k + 1) = (acc + measurement) >> 1

Using this filter results in that new average is the average between the old average
and the new measurement. This is a filter using recursive feedback and thus cate-
gorized as an Infinite Impulse Response filter (IIR-filter). A difference equation for
the filter is presented in Equation 7.2:

y(n) = 0.5 · y(n − 1) + 0.5 · x(n) (7.2)

where y(n) is the new average, y(n-1) is the old average and x(n) is the measure-
ment.

7.6 Strategy to go from the case "slow computer" to

"fast computer"

45
Chapter 7. Results

Using a PWM frequency of 6 KHz and calculating the duty cycle every fourth PWM
period the resulting sample time of duty cycle calculation is calculated by inserting
the values in Equation 6.3:

1
Ts = 4 · TPW M = 4 · = 166.67µs (7.3)
6 · 103 Hz

The time needed to calculated the duty cycle was experimentally determined and
set to 20 % of the maximum count value PER - using Equation 6.4 the time delay is
calculated as:

OFFSET 0.2 · PER

dt = · TPW M = · TPW M = 0.2 · TPW M (7.4)
PER PER

The relative duty cycle calculation delay is then calculated using the Equation 6.5
and inserting the values from calculation 7.3 and 7.4 as:

0.2 · TPW M
= 0.05 (7.5)
4 · TPW M

7.7 Current control with blocked rotor

In the background section we compared Equation 2.9 with Equation 2.10 and setting
e(k) = i∗ (k) − i(k), e(n) = i∗ (n) − i(n) we see that, if we disregard the feed-forward
in Equation 2.9 that these two equations are almost the same. The boundary of the
summation differs with one sample. We identify that K p = TLs + R2 and Ti = L 1 Ts .
R+ 2
Inserting the motor parameters and sampling time with the motor resistance of 1
1
Ohm, motor inductance of 6.9 mH and a sample period of 6KHz = 166.67µs we
get:

L R 6.9 · 10−3 1
Kp = + = −6
+ = 41.9 (7.6)
Ts 2 166.67 · 10 2
1 1
Ti = L Ts = −6 = 143.2 (7.7)
6.9·10−3
R+ 2 1 + 166.67·10
2

Inserting the calculated Ti and K p in expression for the discrete PI-controller

described in Equation 2.10 and recognizing that the output u(k) is the voltage refer-
ence u∗ (k) given in Equation 2.9 we get:
 Ts n=k   166.67 · 10−6 n=k 
u∗ (k) = K p e(k) + ∑ e(n) = 41.9 e(k) + ∑ e(n) (7.8)
Ti n=0 143.2 n=0

46
 7.7 Current control with blocked rotor

Using a supply voltage of 30 V and inserting the calculated voltage reference from
Calculation 7.8 into the algorithm to calculate the duty cycle found in Equation 5.1
and 5.2 we get:
 u∗ (k)
∗
 30 , 0 ≤ u (k) ≤ 30

∗ ∗
D1 (u (k)) = 1, u (k) > 30 (7.9)
u∗ (k) < 0

0,

 u∗ (k)
∗
− 30 , 0 > u (k) ≥ −30

D2 (u∗ (k)) = 1, u∗ (k) < −30 (7.10)
u∗ (k) ≥ 0

0,


If the calculated duty cycle is a value between 0 and 1 then to generate a matching
PWM signal we rewrite Equation 6.2 to get an expression for how to calculate the
compare register based on a given duty cycle - this is shown in Calculation 7.11

D · PER = CCx (7.11)

The final proposed current control algorithm with a discrete PI-controller, tuned
with the motor parameters, able to control the current in both directions can be
found in Appendix A.1. It is based upon combining calculations 7.8, 7.9, 7.10 and
7.11.
Due to time constraint the entire control algorithm covering both PWM signals
was not implemented, this because a solution to handle both varying dead-zones
was not implemented in time. Instead the algorithm was limited to only PWM-
signal, meaning that the supply voltage could only be applied in one direction and
thus only able to control torque in one direction. A calculation error that was not
discovered until the writing of this thesis meant that current control experiments that
had been carried out - had been done without a tuned PI-controller. The untuned PI-
controller had a K p = 126 · 103 and Ti = 7. Figures 7.5 and 7.6 shows the current
control with the simplified untuned PI-controller and a blocked rotor. The data is
logged using the double buffer strategy using a DMA to write to a SD-card, which
is presented in Section 6.6.

47
Chapter 7. Results

Figure 7.5 The Figure shows current measurements captured with the DMA and
saved to the SD-card - with the blue dots representing measured current, red dot
indicating the current reference - with the red line indicating a change in the current
reference from -200 mA to -400 mA.

48
 7.7 Current control with blocked rotor

Figure 7.6 The Figure shows current measurements captured with the DMA and
saved to the SD-card - with the blue dots representing the current reference and the
red lines indicating the measured current.

49
8
Discussion

8.1 Final iteration of current measurement setup

The final current measurement setup is designed to measure bidirectional currents
of up to 16.5 A. However no such experiments were done and the largest current
that was measured was around 10 A - this due to being limited by the amount
of current that the power supply could generate. Noise in the system not related
to the common-mode transients caused by the switching could have been further
investigated. To verify the current measurement setup a third party tool should have
been used to measure the motor current. The measurement done with the third party
tool should then have been compared with the measurement done with the system
designed in this thesis. This to be able to verify the constructed current measurement
setup - this was however not done.

8.2 Dead-zone strategies

It can be argued that the implementation of dead-zone strategies to reduce the noise
generated by common-mode transients effectively worked. This based upon com-
parison of the noise levels shown in Figure 7.4, in which the full strategy to deal
with both the varying and stationary dead-zone had been implemented, with Figure
7.2, where none of the dead-zones issues had been addressed. The implementation
of dead-zone strategies comes at the cost of skipping measurement samples. The
amount of samples that are skipped during a PWM-period is determined by the
duty cycle. Fewer samples are skipped if the varying and stationary dead-zone co-
incides or if there is no switching during the PWM-period. The noise not related
to common-mode transients can be seen in the first section of Figure 7.4 where the
duty cycle is set to zero.
A higher PWM frequency amplifies the problem with noise generated by
common-mode transients due to increased switching. Moreover, increasing the
PWM frequency decreases the number of ADC measurements that can be taken
during a period - for a given sampling rate. Less measuring samples aggravates the

50
 8.3 Current averaging calculations

negative impact of having to skip a fixed amount of samples. This imposes an upper
limit on how fast the PWM frequency can be when using this strategy.
Skipping samples introduces an varying delay in the current measurements that
are used to calculate the next duty cycle. The effect of this potential delay has not
been fully investigated. The implemented control system used an ADC sampling
frequency of 250 KHz, resulting in one sample being taken every 4 µs. In contrast,
the period of a single PWM signal is approximately 41.7 µs, allowing for potentially
10-11 measurements being taken during one PWM period. In the the worst-case
scenario, 6 measurements are skipped during one PWM period when implementing
the strategy to skip both the varying and stationary dead-zones.

8.3 Current averaging calculations

Introducing the averaging filter presented in Calculation 7.1 introduces a delay in
the system. The effect of this filter has not been investigated or included in any mod-
els due to time constraints. The filter is an infinite impulse response filter (IIR-filter)
and the difference equation of the filter is presented in Equation 7.2. It is assumed
that this filter can have a big impact on the control systems ability to control the mo-
tor current and thus it would be of great value to analyze the consequences further.
It would also be a good idea to introduce an upper and lower value on the measure-
ments to reduce the impact of noise - measurements that fall outside this range are
then truncated. This could have helped to reduce the impact of large measurement
noise spikes. Such spikes can be seen in Figure 7.6. This was however not done.

8.4 SD card benchmark evaluation: assessing write

speed and latency
The test was done when the microcontroller ran a benchmark software and not in the
load simulator system, due to this it only indicates the best possible performance.

8.5 Writing to SD card in real-time using DMA

The strategy to write to the SD-card in real time was successful this based on being
able to measure the current used to calculate the duty cycle without generating any
errors - which was the goal. An other way to verify this would have been to have a
simplified known algorithm that was to be recorded using the implemented DMA
strategy and then post-analysing the captured data stored in the SD-card. This last
verification method was however never done.

51
Chapter 8. Discussion

8.6 Strategy to go from the case "slow computer" to

"fast computer"
As discussed in the background chapter the definition of a "fast computer" is when
the computer is able to provide a current reference for a sampling interval that is
based on measurements sampled in the beginning of the sampling interval. This
can be broken up in two parts, first that we make sure that the next duty cycle is
calculated close in time to its use and the second part is that the current measurement
is taken close in time to the calculation. Starting with the delay of when we start to
calculate the duty cycle until when it is put in use. This calculation delay can be seen
in Calculation 7.4 and is calculated to 0.2 · TPW M = 8.3 µs. Relating this calculation
delay to the total sampling period of the duty cycle is done in Calculation 7.5, it
is calculated to be a delay of 5% of the sample period. In the ideal case of a "fast
computer" there would be no delay, however 5 % can be argued to be a small enough
delay such that it can be said that the duty cycle is calculated close enough in time
to its use to be regarded as the case of "fast computer".
Investigating the current measurement delay we recognize that this delay is the
distance from the last current measurement to when the duty cycle is started to be
calculated. The ADC takes a measurement every 4 µs, this means that without any
filtering or averaging a potential delay of up to 4 µs is present. However both a
moving averaging filter and a strategy to skip samples to deal with the dead-zones
were used in the system, which both introduces delays. In the worst-case scenario,
the varying dead-zone strategy is timed in such a way that the next three adjacent
samples, preceding the calculation of the duty cycle, are skipped. This would intro-
duce a delay of up to 16 µs from the last current measurement. Disregarding the
effect of an averaging filter this would mean a combined delay of up to 24.3 µs, cal-
culated as the sum of 12 µs (delay caused by skipping three adjacent samples), 8.3
µs (duty cycle calculation delay) and 4 µs (potential delay until next measurement).
This total delay is approximately 14.6 % of the duty cycle period when disregarding
the effect of the averaging filter and assuming the worse case. The current control
model shown in Figure 4.3 does not compensate for this delay. To reduce the total
delay a faster processor could be used (to reduce the duty cycle calculation delay)
and the sampling rate of the ADC could be increased (to reduce the maximum delay
between measurements). The number of samples to be skipped when dealing with
the dead-zones could be experimented with and perhaps be better tuned to reduce
the delay it introduces - as this was not thoroughly experimented with. The effect of
the delayed caused by the filter was not investigated.

8.7 Current control with blocked rotor

In Section 4.6 challenges with current control of unloaded DC motors is discussed.
To simplify the experiment to control the motor current to be able to follow a current

52
 8.7 Current control with blocked rotor

reference the motor rotor was blocked. This to prevent the motor from accelerating
and changing the dynamics of the motor current by the induced back-emf. It can be
argued that this will be somewhat the case when the control system (the red box) is
used to simulate a load in the system presented in Figure 1.1. This due to the system
motor (blue box) trying to control the connected physical axis to achieve a speed
reference - and thus applying a counter-force to achieve a fixed speed. However this
has not been investigated or experimented with and will require a more in depth
analysis of how it could work.
A PI-controller tuned based on motor parameters and the sample period of the
duty cycle was calculated. However due to not having implemented a way to deal
with the varying dead-zone strategy for both PWM signals a simplified control al-
gorithm was used instead. This algorithm only utilizes one of the half-bridges in
the H-bridge and can thus only apply current (torque) in one direction. Due to a
calculation error the current control experiments had been done with an untuned PI-
controller. The tuned PI-controller has a K p = 41.9 and a Ti = 143.2, the untuned
PI-controller that was used in the experiments has a K p = 126·103 and a Ti = 7. This
is a K p that is approximately 3000 times bigger than the tuned controller, meaning
a very small current error will most likely result in outputting the maximum duty
cycle and completely overtake the control algorithm. Figures 7.5 and 7.6 shows the
experiment of the current control with the untuned PI-controller. Even though the
PI-controller is very badly tuned, it is possible to see that the controller is able to
track the current reference. It would have been very interesting to see the how the
tuned algorithm would perform - however due to time constraints there were no
time to conduct such experiments.

53
9
Conclusion

The project was not a further development on an already existing system and was
designed and built from scratch. A time consuming process, but benefited from
not being locked to certain solutions or components. A lot of work was mainly
conducted to research, to understand and implement basic functionality - such as
how to operate the peripherals of the Arduino MKR Zero.
The goal to create a system that could hardware simulate the stored rotational
energy in a door has not been achieved. This due to no such experiments has been
conducted. Current control of a brushed DC motor was partially achieved which
was assumed to be the foundation to generate a driving force that could be used to
simulate stored rotational energy. Current control was only partially achieved due
to being able to control the current in one direction and not both. A very untuned
PI-controller was used in current control experiments and could follow a current
reference but with heavy oscillation. The oscillations had an amplitude of approxi-
mately 200 mA. It was not established if these oscillations were due to the control
algorithm being untuned or due to measuring noise in the system. A strategy with
double buffers and utilizing a DMA was implemented and tested and was success-
fully able to record the current used in the current control algorithm. The captured
current was stored to a SD-card, so that it later could be analysed.
A current measuring circuit was designed and implemented with the purpose
to be able to measure the inline motor current. The measuring circuit worked on
the principle of measuring the voltage drop across a shunt resistor. The circuit was
designed to be able to handle bidirectional currents of 16.5 A due to this the shunt
resistance had to be small. A current sense amplifier was utilized to amplify the volt-
age drop across the shunt to be able to utilized the full voltage range of the ADC.
The current sense amplifier was also selected due to its ability to reject common-
mode voltage transients. Common-mode voltage transients are created due to the
switching of the H-bridge. The current sense terminals either experience the supply
voltage or ground and the switching of the H-bridge causes both of these terminals
to rapidly change potential. The common-mode transients caused a large distur-
bance in the output of the current sense amplifier and is referred to as a measuring
dead-zone in this thesis. It was analysed that up to two dead-zones could appear

54
 9.1 Limitations

during a PWM-period. The first dead-zone could only occur in the beginning of
the PWM period and due to this was named the stationary dead-zone. The second
dead-zone could occur at any time during the PWM period and was thus named
the varying dead-zone. The time a dead-zone lasted was measured and a strategy to
deal with them was implemented. The strategy was based upon skipping measuring
samples. The stationary dead-zone was dealt with by always skipping the first sam-
ples at the start of a PWM period. The varying dead-zone was dealt with by having
the PWM signal as an input to the microcontroller. A transition in the PWM signal
triggered an interrupt that caused the system to skip samples. This method dealt
with the disturbances caused by the common-mode voltage transients effectively
but came at the cost of introducing a delay in the current measurements.

9.1 Limitations
Due to time constraint the current control was limited to the scenario with blocked
rotor and only using one direction of the H-bridge.

9.2 Future work

The PI-controller needs to be tuned and the step response of the system needs to
be properly investigated and analyzed. Add anti-windup to the integral part of the
PI-regulator. Attach the second PWM-signal to an interrupt pin of the Arduino to
deal with the varying dead-zone from the second PWM signal. Implement the full
proposed algorithm to be able to control the motor current in both directions. Verify
that the algorithm for handling dead-zones works when operating both half-bridges
in the H-bridge. Investigate how the implemented averaging filter effects the current
control.
Improve the speed calculation to include getting the direction of the rotation.
Add the back-emf as a feed forward to extend the PI-controller to a PIF-controller
and then analyse the impact this has on the step response.
Add support to measure the current in one more motor - preferably reuse the
existing solution. Upgrade the microcontroller to contain peripherals with at least
two internal differential ADC (to measure the current in the load/dynamics simula-
tor and the system motor) and preferably two or more DMA. Evaluate the method
of using the DMA to collect current measurement and saving these to a SD-card.
Compare by introducing an external tool to do motor current measurements.
Start to research how the force simulator can be integrated in a HIL-rig. Analysis
which operating loads the load simulator can be operated at and what the limiting
factors for these are.

55
A
Code

A.1 Tuned control algorithm

K p = 41.9, Ti = 143.2, VV M = 30 V and Ts = 166.67 µs

currentNow = (avgVolt * ARDUINO_GAIN_ERROR - noCurrVolt) * 10; // [mA]

currentError = currentRef - currentNow; // [mA]
errorSum = errorSum + currentError; // [mA]
dutyCycle = (currentError * 1.397*10^-3 + 1.625*10^-9*errorSum)

if (dutyCycle => 0)
{
REG_TCC1_CCB0 = 0;
REG_TCC1_CCB1 = min(dutyCycle, 1) * 2000;
X = 1;
Y = 0;
}
else
{
REG_TCC1_CCB0 = -max(dutyCycle, -1) * 2000;
REG_TCC1_CCB1 = 0;
X = 0;
Y = 1;
}

56
 A.2 Simplified and untuned control algorithm

A.2 Simplified and untuned control algorithm

K p = 126 · 103 , Ti = 7, VV M = 30 V and Ts = 166.67 µs
currentNow = (avgVolt * ARDUINO_GAIN_ERROR - noCurrtVolt) * 10; // [mA]
currentError = currentRef - currentNow; // [mA]
errorSum = errorSum + currentError; // [mA]
dutyCycle = (currentError * 4.2 + errorSum/10)

if (dutyCycle => 0)
{
REG_TCC1_CCB0 = 0;
REG_TCC1_CCB1 = min(dutyCycle, 1) * 2000;
X = 1;
Y = 0;
}
else
{
REG_TCC1_CCB0 = 0;
REG_TCC1_CCB1 = 0;
X = 0;
Y = 0;
}

57
Bibliography

Alaküla, M. and P. Karlsson (2011). Power Electronics: Devices, Converter, Control

and Applications. Department of Industrial Electrical Engineering and Automa-
tion.
Bengtsson Lars, 1. (2020). “Implementation of control algorithms in small embed-
ded systems.” Engineering 12:9, pp. 623–639. ISSN: 1947-3931. URL: https:
//ludwig.lub.lu.se/login?url=https://search.ebscohost.com/
login.aspx?direct=true&AuthType=ip,uid&db=edsswe&AN=edsswe.
oai.gup.ub.gu.se.296432&site=eds-live&scope=site.
Bridgmon, J. and C. Andrews (2016). Current sensing for inline motor-control ap-
plications. URL: https://www.ti.com/lit/an/sboa172/sboa172.pdf?
ts=1653421285476. (Accessed: 2022-05-25).
Glad, T. and L. Ljung (2003). Reglerteori - Flervariabla och olinjära metoder. Stu-
dentlitteratur.
Greiman, B. (2021). Sdfat version 2. URL: https : / / github . com / greiman /
SdFat. (Accessed: 2022-06-05).
INA240 (2016). sbos662c. INA240 –4-V to 80-V, Bidirectional, Ultra-Precise Cur-
rent Sense Amplifier With Enhanced PWM Rejection. Texas Instruments. URL:
https : / / www . ti . com / lit / ds / sbos662c / sbos662c . pdf ? ts =
1653987373741. (Accessed: 2022-05-25).
Itarsiwala, R. (n.d.). Getting started with current sense amplifiers: choosing an ap-
propriate shunt resistor. URL: https : / / training . ti . com / getting -
started - current - sense - amplifiers - session - 4 - how - choose -
appropriate - shunt - resistor ? context = 456802 - 1138857 - 456803.
(Accessed: 2022-05-25).
Leonhard, W. (2012). Control of Electrical Drives. Springer Berlin Heidelberg.
ISBN : 9783642976469. URL : https://books.google.se/books?id=O%
5C_XvCAAAQBAJ.

58
 Bibliography

SAM D21 Family Data Sheet (2018). ds40001882d. Low-Power, 32-bit Cortex-
M0+ MCU with Advanced Analog and PWM. Microchip Technology Inc.
URL: https : / / content . arduino . cc / assets / mkr - microchip _
samd21 _ family _ full _ datasheet - ds40001882d . pdf ? _gl = 1 *
12ox7kk * _ga * MTQ4NTMyNzM2Ni4xNjQyNTgxMzU1 * _ga _ NEXN8H46L5 *
MTY1MzQ5ODE4MC4xNTQuMS4xNjUzNDk4MTgyLjU4. (Accessed: 2022-05-25).

59
 Document name
Lund University
MASTER’S THESIS
Department of Automatic Control Date of issue
Box 118 October 2023
SE-221 00 Lund Sweden Document Number
TFRT-6220
Author(s) Supervisor
Eric Schyllert EricWarnquist, ASSA ABLOY Entrance Systems AB
Per Byrhult, ASSA ABLOY Entrance Systems AB
Anton Cervin, Dept. of Automatic Control, Lund
University, Sweden
Karl-Erik Årzén, Dept. of Automatic Control, Lund
University, Sweden (examiner)

Title and subtitle

Dynamic Torque Control of Brushed DC Motors for Hardware-in-the-Loop Integration
Abstract

The main goal with the thesis was to design, construct and evaluate a system able to drive and brake a
mechanical axis to simulate loads and stored rotational energy - a hardware force simulator. The force
simulator was then to be integrated in a existing hardware-in-the-loop testing rig to enhance the
mechanical dynamics of door simulations. The force simulator was to be realised using a brushed DC
motor and by controlling the motor torque. Controlling the motor torque of a brushed DC motor is
achieved by controlling the motor current - and this is mainly what this thesis is about. The thesis
explains the foundations of an embedded system capable to control and log the current of a brushed
DC motor.
The force simulator consisted of a microcontroller, a motor card, a circuit to measure the inline
motor current and a brushed DC motor. The circuit to measure the current was designed and
implemented, consisting of a shunt resistor and a current sense amplifier. The shunt was placed in
series (inline) with the motor. A software strategy was developed and implemented to deal with noise
due to common-mode voltage transients (caused by motor control using PWM). This strategy came
with a cost of introducing a measurement delay in the system.
A discrete PI-controller to control the motor current was researched and implemented. An expression
to optimal tune the controller based on motor parameters and the sampling period was researched.
Current control experiments were conducted but due to a calculation error the PI-controller was very
untuned, this compared to the tuned expression. This error was discovered very late in the work and
there was no time to redo the experiments. The experiments with the implemented untuned
PIcontroller were conducted and a current reference was successfully tracked, however with noise
which had an amplitude of approximately 200 mA.
A strategy to record current measurements used in the current control, without missing any samples
was successfully developed. It worked by using a double buffer system, filling the buffers utilizing a
DMA and then written to a SD card once a buffer was filled. With the main objective to develop the
capability to be able to analyze the current control algorithm.

Keywords

Classification system and/or index terms (if any)

Supplementary bibliographical information

 ISSN and key title ISBN
0280-5316
Language Number of pages Recipient’s notes
English 1-59
Security classification

http://www.control.lth.se/publications/