See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317376629

Electric vehicle power electronics cooling system pump control using fuzzy logic

Article · June 2017

DOI: 10.19101/IJACR.2017.730019

CITATIONS READS
3 1,005

2 authors, including:

Adnan Shaout
University of Michigan-Dearborn
271 PUBLICATIONS 1,295 CITATIONS

SEE PROFILE

All content following this page was uploaded by Adnan Shaout on 14 June 2017.

The user has requested enhancement of the downloaded file.

 International Journal of Advanced Computer Research, Vol 7(31)
ISSN (Print): 2249-7277 ISSN (Online): 2277-7970
Research Article
http://dx.doi.org/10.19101/ IJACR.2017.730019

Electric vehicle power electronics cooling system pump control using fuzzy
logic
Adnan Shaout* and Jason Cooper
Electrical and Computer Engineering, University of Michigan-Dearborn

Received: 16-February-2017; Revised: 17-May-2017; Accepted: 20-May-2017

©2017 ACCENTS

Abstract
Electric vehicles rely exclusively on electrical energy not only for propulsion, but also to power all of the vehicle
subsystems. It then becomes critical that each subsystem is controlled in the most energy efficient manner possible. Some
of the most energy consuming elements of a vehicle are the electric pumps that circulate coolant through an electric
vehicle’s coolant loop. Typically, these pumps are controlled through nothing more than simple temperature switch
points, creating a sink for electrical energy. To improve the overall energy consumption, a fuzzy logic control (FLC)
strategy for controlling an automotive coolant pump has been proposed. The power electronics coolant loop, which can be
found in most hybrid/electric vehicles, have been examined as a temperature-based strategy which is formulated to
control an industry standard pulse-width-modulated (PWM) fluid pump. Simulation of the proposed FLC has shown
improvement of the overall efficiency of the coolant pump and has also reduced the overall energy consumption of the
vehicle-thereby improving the electric propulsion range.

Keywords
Electric vehicle, Fuzzy logic, Power electronics cooling, PWM pump, Temperature control.

1.Introduction Many cooling system strategies, also take into

From safety systems to advanced and alternative account the state/status of the electrical component,
propulsion systems, the evolution of the automobile however this additional layer of control is not
has been as technically advanced as any other relevant to the strategy outlined here as this paper
consumer product. Over the past decade the demand will only detail a temperature-based fuzzy logic
for more electronically controlled mechanical control strategy. For purposes of this paper, a
systems has erupted. Although all of these standard power inverter cooling system will be
Mechatronic systems are required to be functional in evaluated as shown in Figure 1.
almost any environment, nearly all electrical and
mechanical systems are affected by temperature. To Given this hardware setup, a standard automotive
make matters worse for engineers, most electrical and electronic control unit (ECU) is used to control the
mechanical hardware components require a certain pulse width modulated (PWM) coolant pump by
temperature band in order to run optimally. commanding a certain duty cycle. The ECU employs
Thankfully, this temperature dilemma is easily solved a conventional control strategy, which directly
by introducing a cooling system. Most electric associates a crisp input to a crisp output, meaning the
vehicles use multiple coolant loops, assigning one for number of switch points is directly related to the
each component, or system of components. amount of cooling flow obtained. And so, this creates
Fundamentally, a coolant loop consists of multiple a tiered cooling strategy; more simply, the system can
components: (1) component to be cooled, (2) coolant only command two duty cycles: (1) minimum cooling
pump, (3) temperature sensor(s), (4) heat exchanger, duty cycle and (2) maximum cooling duty cycle.
and (6) plumbing. Within such a setup it is necessary Potentially more switch points could be introduced to
that the coolant pump circulates the coolant through create more cooling levels, but this would only create
the hardware. This level of control is almost strictly more discrete levels (with nothing in between). This
dependent upon the temperature of the component to strategy implies that we may get more cooling than
be cooled. what is actually needed. Mainly, this is due to the fact
that the switch points must be defined in such a
*Author for correspondence
111
Adnan Shaout et al.

manner that “one-size-fits-all” for a particular the energy consumption is critical; any energy
cooling level. unnecessarily used by the coolant pump is energy
that is not being used to propel the vehicle, directly
This flaw in the control strategy has two major resulting in a decrease of electric driving range (full
negative implications. First this tiered cooling level electric driving range is the most important feature to
approach means more energy is being used by the a potential customer).
pump than is really necessary. In an electric vehicle

ECU

PWM Coolant Pump

Inverter 1

Inverter 2
Heat Exchanger
Inverter 3

Inverter Temperature
Sensor

Figure 1 Power inverter coolant loop

Next, this approach causes the pump to run at much While there are many ways of implementing a
higher speeds for longer periods of time. This is temperature-based fuzzy logic control system, it will
mainly due to the switch points being designed low be assumed that the hardware setup will be as
enough that higher levels of cooling are required for outlined in Figure 1 and that a standard ECU is used
more scenarios (again the “one-size-fits-all” to control the PWM coolant pump. The fuzzy logic
approach). Consequently the pump running faster for control strategy detailed in this paper has been in the
longer duration’s causes noise, vibration, harshness form of an algorithm that can later be deployed to an
(NVH) concerns. Similar to the first implication, ECU in the form of application level software. It
NVH concerns are magnified on electric drive should also be noted that all Simulink models are for
vehicles due to the removal of noise and vibration simulation purposes only, however they can be used
supplied by the internal combustion engine (ICE). for code/software generation, or as a building block
for more complex algorithms.
One of the best ways to resolve these two concerns is
to create a smoother, more efficient pump control 2.Literature review
strategy – this is where fuzzy logic techniques come While many papers have been written about the use
into play. Through the use of fuzzy logic, it becomes of fuzzy techniques to control temperature-based
possible to use general linguistic terms such as “cold, systems [2, 3, 4], or automotive thermal management
“warm”, and “hot”- as opposed to conventional systems [5], it is the use of this control strategy of a
control, which only has crisp on/off switch points [1]. coolant pump for a power electronics cooling system
Such a fuzzy logic control strategy will then offer in an electric vehicle that makes it state-of-the-art
more even pump control, thereby eliminating the [6,7]. Furthermore, even though a fuzzy control
tiered cooling structure as seen with conventional strategy was devised for an automotive cooling
control methods. In-turn this will greatly reduce the pump, this strategy was investigated from an energy
energy consumption of our system as well as greatly efficiency standpoint – and not necessarily a thermal
reduce the NVH concerns. management perspective. By formulating this fuzzy-
based control strategy for use on an electric vehicle
112
 International Journal of Advanced Computer Research, Vol 7(31)

we are able to reduce the overall energy consumption output a crisp duty cycle that will be used to
significantly (the exact numbers could be found command the pump. It is really the control system–
through simple experimentation)-but, the truly novel more specifically the techniques used within that
part of this idea is that the energy efficiency can now control system-that differentiates conventional
be multiplied by the number of subsystems that control of fuzzy logic based control [8].
employ this strategy (typically electric vehicles
employ a minimum of two cooling loops). When all A. Conventional crisp control methods
of these energy consumption numbers are put The conventional pump control method currently
together, it should equal many miles of additional employed by this system is quite simple – really it is
vehicle propulsion range-which in today’s based on the concept of crisp switching points, which
automotive market addresses the most critical aspect come directly from the temperature inputs (or rather
of electric vehicle production. the maximum temperature of the three temperature
inputs. Even though three temperature readings are
3.Control methods being fed into the control system, for purposes of
From a control strategy perspective the power cooling, i.e. decreasing the temperature of the
electronics cooling system illustrated in Figure 1 can inverters, it is not necessary to use all three
really be broken down into two simple parts: (1) temperatures). The maximum temperature is directly
Inputs: Three inverter temperatures, as read from the compared to two pre-defined switch points: (1)
three appropriately placed temperature sensors, (2) maximum cooling and (2) minimum Cooling. If the
Output: Commanded Duty Cycle (see Figure 2). max temperature is above the minimum cooling to
switch point (currently set to 35°C) then the
corresponding duty cycle will be commanded
Temperature 1 Temperature 2 Temperature 3 (currently set to 50%). Similarly, if the max
(°C) (°C) (°C) temperature is above the maximum cooling switch
point (50°C) then the corresponding duty cycle will
Max be commanded (90%).

An additional switch point is defined for each cooling

level. For minimum cooling (32°C), this switch point
Control System turns the pump off (50% to 0%), and for maximum
cooling (47°C) this switch point commands the pump
to the low duty cycle (90% to 50%). Figure 3
illustrates a simple Simulink model of the described
Commanded Duty conventional crisp control method. When this
Cycle (%) simulation is run (see Figure 4), we can clearly see
Figure 2 Inputs/outputs for inverter cooling system that the commanded pump duty cycle is directly
dependent upon the temperatures being read. The
From Figure 2 it is now obvious that the inputs and afore mentioned tiered cooling levels are also clearly
outputs to this system (or really any system) will be visible. These jumps in the commanded duty cycle
common between the conventional control strategy are what create the inefficiencies-again causing
and the fuzzy logic control strategy. Both will have unnecessary energy usage.
crisp temperature values as their input, and both will

Figure 3 Simulink model for conventional crisp control

113
Adnan Shaout et al.

50% DC (min cooling) 50% DC (min cooling)

90% DC (max cooling)

35°C
35°C

47°C
50°C
32°C 32°C

Figure 4 Pump duty cycle command (top), max temperature reading (bottom)

B. Fuzzy control methods coolant levels, we can lower the final duty cycle
Fuzzy logic will allow us to generalize the command and thereby lowering the energy
conventional crisp control methods outlined in the consumption of the coolant pump.
previous section by creating non-sharp boundaries
between the switch points. This will provide the Generically the fuzzy control system algorithm will
flexibility needed to eliminate the duty cycle tiers consist of three main parts (Figure 5) [8, 9]:
visible in Figure 4. In-turn by smoothing out the

Membership Membership
Function Function

Crisp Crisp
Fuzzification Defuzzificaion
Inputs Ruleset Outputs

Fuzzy Inputs Fuzzy Outputs

Inference
Engine

Figure 5 Block diagram of fuzzy logic algorithm

Fuzzification: The inputs are first compared to the Defuzzification: The fuzzy output is converted to a
input membership function. This allows the control crisp value that the ECU can demand of a system.
algorithm to determine the degree to which the
linguistic variable is true. a. Crisp input/output system interfaces
Inference Engine: Each rule in the fuzzy ruleset is A fuzzy logic controller (FLC) really refers to the use
then processed with the fuzzy input value. This of fuzzy logic within the control strategy itself, this
process maps the input to a corresponding fuzzy then implies that the inputs, and subsequently the
linguistic output variable. Given the current input outputs, of the system will be identical to their
conditions, this phase of the algorithm will provide a conventional counterparts (we live in a crisp world!)
system output. [10]. In Figure 2 the crisp input and output variables
114
 International Journal of Advanced Computer Research, Vol 7(31)

were defined for use in both the conventional control the next function. This technique is really a key to
methods as well as the fuzzy control methods -Table using fuzzy logic.
1 again outlines the inputs and output of our system.
As was the case for conventional control methods, The following equations have been used to find the
the power inverter cooling system inputs can really membership function values for each of the linguistic
be simplified to one input. values:

Table 1 Inputs / outputs from the inverter coolant

system { (1)
;
Input Input type Input unit
Temperature #1 °C
Temperature #2 °C
Temperature #3 °C
Output Output Type Output Unit
;
(2)
Duty Cycle %
;
This is true since the three inputs were all {
temperature and all temperatures were part of the
same cooling loop – more simply, only the maximum
temperature of a cooling system is used since it will ;
(3)
dictate when cooling is required. Again, as before,
only the maximum temperature will be taken into ;
account for the pump control strategy. This will allow {
us to define only one input variable.

The fuzzification of the input variable will start by ;

(4)
defining it as a linguistic variable. In this case, the
input variable will linguistically be denoted as ;
temperature. The next step in the fuzzification {
process is creating a membership function of
temperature. By partitioning it into five
; (5)
memberships, function envelopes, we can create {
overlapping regions, as seen in Figure 6. Each
function will then represent a linguistic value for the ;
linguistic variable. In this case the following
The universe of discourse was found through
linguistic values have been defined for temperature:
experimental data and is really associated with an
cold, cool, warm, hot, and very hot.
appropriate temperature range (taking into account
μ the data and values associated with their crisp switch
Cold Cool Warm Hot Very Hot point counterparts).
1
Universe of Discourse: * +
A similar fuzzification strategy was performed on the
crisp output. In this case, of course, the fuzzified
output will be used to map the input to the output–
0
30 35 40 45 50 after-which it will be defuzzified to ultimately
Temperature °C provide a crisp output duty cycle. In this system we
have one output variable, which linguistically will be
Figure 6 Membership function of linguistic input
denoted as duty cycle.
variable temperature
Figure 7 shows the linguistic values which have been
By overlapping the partitions, the linguistic input
defined for duty cycle for very low, low, medium,
temperature now has a smooth transition between
high, and very high.
states – more simply put, as the temperature changes,
its membership value in one function begins to
Universe of Discourse: * +
decrease while it increases its membership value in
115
 Adnan Shaout et al.

μ The rule base designed for this scenario is as follows:

Very Low Low Med High Very High 1. IF temperature is Cold, THEN duty cycle is Very
1
Low.
2. IF temperature is Cool, THEN duty cycle is Low.
3. IF temperature is Warm, THEN duty cycle is
Medium.
0 4. IF temperature is Hot, THEN duty cycle is High.
10 30 50 70 90 5. IF temperature is Very Hot, THEN duty cycle is
Duty Cycle %
Figure 7 Membership function for linguistic input Very High.
variable duty cycle
With this ruleset, we can clearly see the connection
The following equations have been used to derive the between the input variable and the output variable
output membership functions. (both fuzzy). At each instance of time a temperature
reading will intersect the input membership function
and then be mapped to the output membership
function (as seen in Figure 7) [10].
{ (6)
;
The analysis of Figures 6 and 7 shows that the
membership functions of the linguistic input variable,
; temperature, and the linguistic output variable, duty
cycle, are identical in shape-this was done
(7)
intentionally. By defining duty cycle to have the exact
; same number of linguistic values as temperature we
{ are able to ensure a one-to-one mapping between the
input and output. More simply, the membership
; function value of the temperature is equal to the
membership function of the duty cycle.
(8)
; c. Defuzzification
{ After the rule base has been assessed we are left with
a linguistic output value and a membership function
; value, however, since there is a one-to-one
(9) relationship between the input and output, we can
simply use the membership function and the current
; output state to find the crisp duty cycle output. This
{ crisp output will be used to command the cooling
system pump.
;
(10) The input to output membership function mapping,
{ illustrated in Figure 8, clearly shows that the
; membership function value of the input is equal to
the membership function value of the output. This
implies that in order to defuzzify the output, we can
b. Inference engine / rule assessment simply use the input membership function values (as
By defining a set of rules, the fuzzy input can be found from equations 1-5 to find the crisp duty cycle
mapped to the decision that the ECU will need to from Equations 6-10, as seen in Table 2.
make in order to properly control the pump (i.e. the
fuzzy output). Typically in fuzzy logic a simple The exceptions are a temperature input less than
format is used when formulating the rule set: “If X 30°C, or greater than 50°C. Due to hardware
Then Y “.This format is also very convenient for the limitations the coolant pump cannot be commanded
ECU that will ultimately interpret these rules to make below 10%, or beyond a 90% duty cycle, therefore
a decision on the output [1]. Each state, or linguistic any temperature below 30°C will default to a 0% DC
value, of the linguistic input and linguistic output and any temperature above 50°C will default to a
variables will then have a rule associated with it.

116
 International Journal of Advanced Computer Research, Vol 7(31)

90% duty cycle (which can be seen in the Simulink the fuzziness of the output, this technique will yield
model, as well as Figure 8 and Table 2). the same crisp value as using the higher cooling
request (the slope equation simply changes, as seen in
It should also be noted that when multiple rules are Equations 1-5 and 6-10. This technique is mainly
activated, the rule with the lowest cooling request used for simplicity and modelling purposes.
will be used to calculate the crisp duty cycle. Due to

Very
Cold Cool Warm Hot Very Hot Low Med High Very High
Low
1 1

0 0
30 35 40 45 50 10 30 50 70 90

15 31 35 42 0 14 30 58

Figure 8 Mapping of a crisp temperature to a crisp duty cycle using fuzzy techniques (5 Time Samples)

Table 2 Defuzzification equations

Temperature range (°C) Linguistic value Defuzzification equation / value
Cold / Very Low 0
Cold / Very Low , ( ) -
Cool / Low , ( ) -
Warm / Medium , ( ) -
Hot / High , ( ) -
Very Hot / Very High 90

4.Simulation of fuzzy control methods by each of the three temperature sensors). Each
In order to evaluate the performance of the fuzzy temperature intersects the fuzzy input membership
control strategy outlined in this paper, a Simulink function at least once, but typically twice. From these
model was created (see Figure 9). Simulink was intersection points corresponding membership
chosen as the modeling tool for two main reasons: (1) function values were found for each linguistic value.
it allows a user to fully model a system, which can be Through the rule base, each input is then mapped to a
easily read and interpreted by outside uses, and (2) corresponding fuzzy output, which is then defuzzified
since Simulink can auto-generate controller level using the formulas seen in the defuzzification section.
code, has such a model becomes useful for
implementing a control strategy in an ECU using Table 3 Crisp temperature input values and crisp
software (typically C code). duty cycle output values (at each instance of time)
Time Temperature (°C) Duty cycle (%)
Similar to the simulation run using a conventional Sample
control model (Figure 3), the temperature inputs for 0 15 0
1 31 14
the fuzzy numerical example/simulation will be 2 42 58
cycled through the values seen in Table 3. Given 3 35 30
these system inputs, a duty cycle command is found 4 31 14
using the control strategy outlined in the preceding 5 47 78
parts of this paper. Figure 10 and Table 3 outline the 6 52 90
simulation results. 7 47 78
8 30 10
At every time sample the temperature value indicated 9 36 34
is fed into the control system (as though it was read 10 20 0

117
Adnan Shaout et al.

Figure 9 Simulink model for temperature-based fuzzy logic pump control

118
 International Journal of Advanced Computer Research, Vol 7(31)

58% DC
10% DC
34% DC
14% DC 14% DC

90% DC

0 Time 1 2 3 4 5 6 7 8 9 10
Figure 10 Duty cycle command (top), max temperature reading (bottom) for fuzzy controls

Figure 11 Comparison of simulation using conventional (purple) and fuzzy control methods (yellow)

5.Conventional control methods vs. fuzzy with the fuzzy control strategy–this is a direct
control methods byproduct of using fuzzy logic. With further
If we compare Figure 4 to Figure 10 (see Figure 11), experimentation and tuning of the membership
it is very evident that the fuzzy control methods functions and the temperature partitions used for each
designed here were successful. The commanded duty linguistic value of temperature, these regions could
cycle is much smoother and much more efficient be easily improved–and the reduction in energy
when fuzzy logic techniques were employed in the consumption could be further improved.
control system. With the elimination of the tiered
cooling levels, comes the improvement in NVH – the 6.Conclusion
biggest contributor to noise and vibration is a pump Through the use of fuzzy logic techniques a control
that switches on and off. Furthermore, we can see method was developed for use on the power
that there are durations of time where the pumps are electronics cooling system in an electric vehicle. This
running at lower duty cycles, which of course can be control strategy can not only improve the overall
directly related to the electrical energy/current efficiency of the coolant pump, but it can potentially
consumption of the pumps. Unfortunately, there are reduce the overall energy consumption of the vehicle
also regions where the duty cycle is actually higher -thereby improving the electric propulsion range. The
119
 Adnan Shaout et al.

NVH concerns of an electric vehicle are also [8] Isizoh AN, Okide SO, Anazia AE, Ogu CD.
addressed through the use of this strategy, which can Temperature control system using fuzzy logic
be improved by eliminating the tiered cooling levels technique. International Journal of Advanced Research
seen with the conventional control strategy. As seen in Artificial Intelligence. 2012; 1(1):27-31.
[9] Hanamane MD, Mudholkar RR, Jadhav BT, Sawant
through a Simulink simulation, the proposed fuzzy- SR. Implementation of fuzzy temperature control
based control strategy was able to successfully using microprocessor.
smooth out the commanded duty cycle. Overall, the [10] Aguilar RM, Muñoz V, Callero Y. Control application
control strategy outlined in this paper demonstrated using fuzzy logic: Design of a fuzzy temperature
that it can provide a very practical solution to the two controller. INTECH open access publisher 2012.
main issues associated with the power electronics
coolant pump with an electric vehicle. Dr. Adnan Shaout is a full professor
and a Fulbright Scholar in the
Acknowledgment Computer Science Department at the
None. Electrical and Computer Engineering
Department at the University of
Michigan-Dearborn. At present, he
Conflicts of interest teaches courses in AI, Embedded
The authors have no conflicts of interest to declare.
Systems, Software Engineering,
Computer Architecture, Cloud Computing, Fuzzy Logic
References and Engineering Applications and Computer Hardware
[1] Singhala P, Shah D, Patel B. Temperature control Design. His current research is in applications of software
using fuzzy logic. International Journal of engineering methods, cloud computing, embedded systems,
Instrumentation and Control Systems. 2014; 4(1): 1- fuzzy systems, real time systems and artificial intelligence.
10. Dr. Shaout has more than 34 years of experience in
[2] Chen Z, Mi CC. An adaptive online energy teaching and conducting research in the Computer Science,
management controller for power-split HEV based on Electrical and Computer Engineering fields at Syracuse
dynamic programming and fuzzy logic. In vehicle University and the University of Michigan - Dearborn. Dr.
power and propulsion conference 2009 (pp. 335-9). Shaout has published over 210 papers in topics related to
IEEE. Computer Science, Electrical and Computer Engineering
[3] O'Keefe M, Bennion K. A comparison of hybrid fields. Dr. Shaout has obtained his B.S.c, M.S. and Ph.D.
electric vehicle power electronics cooling options. In in Computer Engineering from Syracuse University,
vehicle power and propulsion conference 2007 (pp. Syracuse, NY, in 1982, 1983, 1987, respectively.
116-23). IEEE. Email: [email protected]
[4] Ning P, Liang Z, Wang F. Power module and cooling
system thermal performance evaluation for HEV Jason Cooper is a full time algorithm
application. IEEE Journal of Emerging and Selected and software engineer at General
Topics in Power Electronics. 2014; 2(3):487-95. Motors Company. Currently, his
[5] Shao C, Sun X, Wang G, Yue Y, Li X, Yang L. assignments include the design and
Numerical modeling and simulation analysis of development of algorithms and
vehicle thermal management system for a power-split software for electrification thermal
hybrid electric vehicle. In international conference on management systems, as well as next-
electrical systems for aircraft, railway, ship propulsion Gen engine thermal management
and road vehicles (ESARS) 2015 (pp. 1-6). IEEE. systems. Jason has over nine years of experience working
[6] Ibrahim BS, Aziah MA, Ahmad S, Akmeliawati R, the in automotive industry–this work has included the
Nizam HM, Muthalif AG, et al. Fuzzy-based development of service diagnostics for engine controls, the
temperature and humidity control for HV AC of design and release of ECU electronics, as well as the
electric vehicle. Procedia Engineering. 2012; 41:904- development of control algorithms. Jason has obtained a
10. B.Sc. in Electrical Engineering from Michigan State
[7] Al-Jarrah A, Salah M, Al-Tamimi A. Fuzzy logic University, East Lansing, MI, in 2007. He is currently a
control design for advanced vehicle thermal graduate student in the Department of Electrical and
management system. In proceedings of the IASTED Computer Engineering at the University of Michigan-
international conference on control and applications Dearborn. Jason is pursuing a M.Sc. in Electrical
2011 (pp. 200-4). Engineering.

120

View publication stats