Applied Energy 118 (2014) 166–172

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Accurate simulation of thermoelectric power generating systems

Andrea Montecucco ⇑, Andrew R. Knox
School of Engineering, College of Science and Engineering, University of Glasgow, UK

h i g h l i g h t s

 We model the thermal and electrical dynamics of thermoelectric power generating systems.
 Both transient and steady-state conditions are considered.
 We develop a computer program for transient simulations of thermoelectric systems.
 The program simulates the electro-thermal coupled effects that occur during changes in the operating conditions.
 Comparison of experimental and simulation results shows great accuracy.

a r t i c l e i n f o a b s t r a c t

Article history: Recent interest in the use of thermoelectric generators (TEGs) to recover waste heat in large-scale appli-
Received 9 October 2013 cations calls for precise simulation to appropriately design complicated and dynamic systems. The aim of
Received in revised form 14 December 2013 this work is to develop a computer tool to accurately simulate the thermal and electrical dynamics of a
Accepted 18 December 2013
real thermoelectric (TE) power generating system.
Available online 14 January 2014
The computer-aided model presented here is able to accurately simulate the non-linear electro-ther-
mal coupled effects which occur during changes in the operating conditions, e.g. temperature or load
Keywords:
changes.
Thermoelectric
TEG
Simulation results are compared to experimental data obtained from a real TE system. The comparison
Heat transfer shows great accuracy both during transients and in the steady-state, thus validating the model as a reli-
Computer simulation able tool to simulate TE generating systems.
Peltier Ó 2013 Elsevier Ltd. All rights reserved.
Seebeck

1. Introduction where p (V) is the Peltier coefficient and the last equivalence comes
from the Kelvin relationship; T j is the junction temperature.
Thermoelectric (TE) devices can directly convert thermal energy Other important phenomena occurring in thermoelectric de-
into electrical energy and vice versa. Thermoelectricity, based on vices are the well-known Joule heating and the Thomson effect.
the fact that charge carriers can be set in motion by a difference The latter states that there is reversible absorption or liberation
of temperature, generally refers to two main physical phenomena: of heat (in excess of the joule dissipation I2 R) in a homogeneous
the Seebeck effect and the Peltier effect. The former states that a material simultaneously exposed to temperature gradient and
certain open-circuit voltage is created in a material kept between electric current:
two different temperatures. The Seebeck coefficient a (V/K) is a
material property that relates the open-circuit thermoelectric po- P T ¼ s I DT ð3Þ
tential V OC (V) with the temperature difference DT (K), or where s (V/K) is the Thomson coefficient, defined as
V OC ¼ aDT ð1Þ
da
The Peltier effect states that a direct current I(A) passing
s ¼ T AVG ð4Þ
dT
through a circuit of dissimilar materials pumps thermal power
The Thomson effect is usually much smaller than the Joule heat-
from one material to the other:
ing [1,2] and ts contribute can be significant under large tempera-
PP ¼ pI ¼ aIT j ð2Þ ture differences [3,4]; also, the Thomson coefficient is difficult to
obtain experimentally, therefore it is often neglected in literature.
Its effect is not included in this article.
⇑ Corresponding author. Tel.: +44 1413302768. Doped semiconductors prove to be the materials with the
E-mail address: [email protected] (A. Montecucco). best thermoelectric properties. Multiple pellets of p- and

0306-2619/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apenergy.2013.12.028
 A. Montecucco, A.R. Knox / Applied Energy 118 (2014) 166–172 167

n-semiconductor are connected electrically in series and thermally cannot be successfully used to accurately simulate the electro-
in parallel to achieve and/or sustain higher voltages, thus forming a thermal coupled effects which take place during changes in the
TE module. This paper will focus on TE generators (TEGs); Fig. 1 system operating conditions, e.g., temperature, power or load
shows a 3D model of a TEG module where thermal energy is ap- changes. In literature some research has focused on this issue;
plied on the bottom (‘‘hot’’ side). The pellets are electrically ser- Lineykin and Ben-Yaakov [19] have divided the TE module into a
ies-connected by solder and an Aluminium-based ceramic layer grid of spatially discretized thermal circuits then transformed into
serves as electrical isolation and mechanical substrate. The result- their electrical analogies. The transient term is included in the
ing module is quite robust and reliable, and operates without any model as a parallel electrical capacitor to take into account an addi-
vibration or noise. It is commercially produced in a wide range of tional time-related term due to the change in stored heat energy. A
sizes from a few millimetres to several centimetres on a side. Mul- similar approach has been followed by Chen et al. [20]. The accu-
tiple modules can be electrically connected in series or parallel in racy of the simulation is related to the number of cells used and
order to achieve higher output voltages and currents. Heat is con- the parameters are difficult to obtain from manufacturers’ data
ducted through the module while additional heat is generated sheets, but most importantly they do not offer a theoretical solu-
(Joule) inside the module and pumped (Peltier) through the hot tion to the problem because their governing equations are based
and cold sides as described by the coloured arrows. In TE power on steady-state solutions. A more appropriate approach would be
generation the Peltier effect is parasitic because it reduces the tem- to study the transient physical equations which describe TE de-
perature difference across the device, thus increasing the module’s vices behaviour; among these the most important is the heat equa-
effective thermal conductivity. Thermoelectric heat pumps (cool- tion. Al-Nimr et al. [21] have already explored this possibility but
ers or heaters) have been used for many years in applications rang- they used fixed temperatures as boundary conditions at the two
ing from IC microcoolers, to refrigerators [5], to power stations [6], sides of the TE module, assuming that those temperatures are
and they are often referred to as Peltier devices. Due to their low not varying, i.e., supposing thermal isolation (or steady-state).
efficiency, often less than 5%, thermoelectric generators have on However, during thermal transients there is exchange of heat
the contrary been used predominantly in military and aerospace through the sides.
projects or for applications in which cost is not as important as Very recently two very interesting works by Cheng and Huang
the ability to reliably generate power in hostile or maintenance- [22] and Meng et al. [23] proposed two models for thermoelectric
free environments. However, interest in TEGs has recently in- coolers. The former slightly overestimates the temperature differ-
creased because of concerns about climate change, coupled with ence in steady-state, while the second has a maximum error of
increasing performance and lower module cost. TEGs can effec- 4.5 K over a temperature difference of around 37 K (with a current
tively lend themselves to sustainable applications of waste heat input of 1 A).
recovery in which thermal energy is rejected to ambient and is We have already provided a mathematical solution of the one-
effectively free, e.g. in vehicles [7–10] and stoves [11,12]. TEGs dimensional heat conduction equation for TE devices that includes
can also be used to harvest geothermal energy [13] or combined internal Joule heat generation and dynamic exchanges of heat
to PV, solar thermal or thermophotovoltaic systems [14–16]. More- through the hot and cold sides in [24].
over, recent advances in TE materials and ‘mass-production’ vol- The aim of this work is to couple the aforementioned solution
umes [17,18] will continue to lead to a further improvement of with the other thermal and electrical phenomena occurring in real
TEGs’ efficiency and reduction of their cost, respectively. TE systems. The resulting physical model, described in Section 2,
It is of fundamental importance to carefully design large-scale takes into account the dynamic relations between the several ther-
systems in which materials cost is of great relevance. However mal masses and the most important thermoelectric phenomena
TE systems are usually composed of heat masses, TEGs and power occurring in a generalised TE system.
and control electronics and they are influenced by several thermal The physical model is then used as the basic structure to devel-
and electronic phenomena whose interaction is complex. More- op a computer tool, described in Section 3, capable of accurate sim-
over TEGs are often employed in dynamic environments which fre- ulations of the thermal and electrical dynamics of a physical TE
quently undergo thermal transients. Actual CAD tools do not yet power generating system. The model is created in Simulink and
include the ability to model thermoelectric effects therefore they Matlab and a comparison between experimental and simulated

Fig. 1. 3D model of a thermoelectric generator showing the main physical effects.

168 A. Montecucco, A.R. Knox / Applied Energy 118 (2014) 166–172

results is presented in Section 4 to demonstrate the effectiveness where mC (J/K) is the heat capacity of the thermal mass, which de-
and accuracy of the proposed simulation model. fines its ability to keep heat energy.
As indicated in Fig. 2, T 1H is used in this paper for the temper-
2. Generic model of a thermoelectric system ature of the hot thermal mass and T H is the temperature on the hot
side of the TEG, while T 1C and T C are the temperatures of the cold
Generally speaking, a TE system is composed of several thermal thermal mass and TEG cold side respectively.
masses which store and exchange heat through conduction and/or
convection. Heat is partially converted to or from electricity inside
3. Computer model
the TE device and/or in other elements, e.g., electrical heaters. It is
very important to optimally design the thermal interconnection of
The solution to the heat equation (HE) presented in [24] was
the TE device with the rest of the system to guarantee optimal per-
programmed in Matlab, therefore the Mathworks suite was chosen
formance [25]. This paper ultimately aims at developing a model
to develop the simulation program. The model includes several
that can be used to aid TE system design.
blocks and performs time-step simulations. Simulink is an excel-
Fig. 2 shows the architecture of a generic TE power generating
lent choice of environment for this task.
system. The TEG is usually in contact with a thermal mass on both
Section 3.2 presents the heat equation block, Section 3.3 deals
the hot and cold side. Power is provided to or removed from the
with the thermal part and 3.4 explains the electrical part of the
thermal masses in the form of electrical or heat power, leading
system.
to changes in the thermal energy stored inside the thermal masses.
Through conduction or convection part of this energy is transferred
to and from the TEG module. The TEG is modelled considering the 3.1. System architecture
sides of the TEG separately from the inner part. The heat equation
(HE) deals with both the heat conduction and generation (Joule Fig. 3 shows the architecture of the Simulink model developed
heating) inside the TEG. Additional heat is pumped at the sides in this work which will be described shortly.
where there is a junction of two dissimilar materials (Peltier The HE computes the transient solution starting from an initial
effect). condition of T C and T H , and with constant values for T 1H ; T 1C and
Part of the energy flowing through the TEG is converted into the load current Iload . It can provide the temperature distribution
electrical power and the process is closely related to the thermo- inside the TE device at any instant in time, even if for this model,
electric effects described by the Eqs. (1) and (2). at system level, we are interested in the temperatures only at the
Conduction and convection equations provide the rate of ther- hot and cold sides of the TEG module. The HE allows the side tem-
mal energy flowing from one medium to the other in time. The peratures T H and T C to change depending on the temperatures T 1H
thermal conduction equation (Fourier’s law) is written in time- and T 1C of the thermal masses in contact with the TEG through
derivative form as: conduction or convection media. The temperature difference
across these media changes dynamically, thanks to the Newton’s
dEcond dT
¼ jA ð5Þ Law of Cooling, as already described in [24].
dt dx In the solution of the HE the temperatures of the thermal
where Econd (J) is the thermal energy, j (W/mK) is the thermal con- masses T 1H;C are considered constant; however, in a real system
duction coefficient, A (m2) is the surface of heat exchange and x (m) during a transient they vary depending on the energy added or re-
is the direction of heat transfer, perpendicular to A. The heat con- moved by Pin and Pout . Similarly, the temperatures at the sides of
vection equation is the TEG are influenced also by the actual load current which in
turn determines the value of the Peltier effect (see Eq. 2). This is
dEconv
¼ hADTðtÞ ð6Þ why in Fig. 3 the ‘HE Block’ is connected to both to the ‘Electrical
dt
Block’ and to the ‘Thermal Block’. The thermal block requires a
where h(W/m2 K) is the heat convection coefficient and and Iload form the electrical block to deal with Peltier effect. The
DTðtÞ ¼ TðtÞ  T 1 is time-dependent. transient evolution of the temperature difference across the TEG
Any lumped thermal mass changes its temperature when device and the system’s dynamic exchange of heat due to P in and
receiving or losing thermal energy: Pout continuously influence the temperatures in the whole system.
Etherm ¼ mC DTðtÞ ð7Þ Therefore in order to couple the continuous-time transient
solution provided by the HE to the temperatures’ evolution in

Fig. 2. Architecture of a thermoelectric power generating system.

 A. Montecucco, A.R. Knox / Applied Energy 118 (2014) 166–172 169

 Electrical parameters of the TEG: internal resistance Rint and

current Iload .
 Temperatures at the beginning of the iteration: T 1H ; T 1C ; T H and
TC .
 Time step duration t step .

bC and bH may vary if the materials in contact with the cold and
hot sides change properties, e.g. increased convection.  can vary
with temperature, but in this model it is considered constant.
The time step size t step is constant and chosen based on the knowl-
edge of the TE system; it is a compromise between the thermal
time constant of the transient evolution inside the TEG and the
temperature change in the thermal masses of the system. Values
between 1 s and 3 s are usually chosen to satisfactorily model
the physical system. For systems with a very high energy flux,
e.g., large engine exhaust gas systems, values of under 1 s might
be more appropriate.
T 1H ; T 1H ; T C ; T H ; Iload ; Rint are fed iteratively at the beginning of
Fig. 3. Block architecture of the presented model. The inputs to the different blocks
every iteration from the rest of the Simulink model.
are written inside the arrows.
As already explained in Section 3.1, the HE block produces two
outputs, T C and T H , that are the temperatures on the cold and hot
the remaining parts of the system (thermal masses, input/output sides after that particular iteration cycle of duration tstep . T C and T H
thermal energy and electrical load), the model is designed in the are then passed through a memory block so that they are effec-
discrete-time domain. Within each time step of duration t step (s) tively available only after tstep , i.e., at the beginning of the succes-
the HE computes the dynamics inside the TEG for a transient of sive iteration. The block could be set to calculate temperatures
duration t step . The temperatures in the remaining parts of the sys- inside the TEG too, but these are not of interest to the aim of the
tem change as described by Eqs. (5)–(7), in which the derivative simulation (nor, usually, to the system designer).
term is substituted by slope terms. This direct examination in dis- If there are TEG modules connected in series or parallel some of
crete terms aids the comprehension of the underlying physical ef- the inputs required by the HE will be scaled accordingly depending
fects in the system. on the total number of modules N tot : Atot ¼ N tot A, while Rint and Iload
To understand how the model is structured, let’s consider two will be provided by the electrical block, as described in Section 3.4.
successive time steps, corresponding at two program’s discrete iter-
ations, called i and i þ 1. It must be noted that the calculations made
3.3. The thermal block
during each iteration are computed directly using the equations
provided in this paper, without any trial-and-error or iterative
The most important tasks executed by the hot side block are the
method. During the iteration i the heat equation uses the electrical
update of the temperature T 1H of the hot thermal mass and the up-
parameters and the heat masses’ temperatures T 1HðiÞ and T 1CðiÞ ,
date of the TEG hot side temperature T H , accounting for the Peltier
considered constant during tstep , and the hot and cold side temper-
effect.
atures T HðiÞ and T CðiÞ as initial values; it computes the values of T Hðiþ1Þ
Consider now the input power transferred to the hot thermal
and T Cðiþ1Þ corresponding to their evolution after the transient time
mass. This can be provided by an electrical heater or by thermal
t step and outputs them through a memory block of duration t step , so
conduction/convection. For the electrical case it is easily measured
that they are actually available only at the start of iteration i þ 1.
while in the thermal case it can be calculated using the well-
The HE block is described in Section 3.2. At iteration i the model also
known laws of heat conduction or convection. As shown in Fig. 3,
updates the temperatures of the thermal masses depending on the
part of the input power goes into changing the quantity of heat en-
quantity of heat energy provided by P in and P out during t step , so that
ergy stored in the hot thermal mass, while the remaining thermal
they are ready for iteration i þ 1. The equations used to update
energy is transferred (through the hot thermal mass) to the TEG.
these temperatures are described in the Section 3.3.
Heat losses can be accounted for in another term P lossH . It can be
written:
3.2. The TEG heat equation block
mH C H DT 1H
Pin ¼ þ hH ADT H þ PlossH ð8Þ
The solution to the HE provided in [24] treats the TEG module as t step
a whole block, i.e. it does not deal with the different materials
where mH C H is the hot thermal mass heat capacity J/K, DT 1H is the
(ceramic, semiconductor, solder) separately; it also does not divide
change in temperature of the hot thermal mass during t step ; hH is the
the semiconductor materials into single pellets or thermocouples.
convection coefficient (or for conduction the conduction coefficient
This is done by design because the parameters for the model are
over the thickness) of the medium between the hot thermal mass
obtained only from the physical dimensions of the module and
and the TEG, A is the area of the TEG and DT H is the temperature dif-
from its direct use in the system, i.e. by representing a real system.
ference across the medium.
The HE code is programmed into a Matlab embedded equation
From Eq. (8) it would be possible to obtain the new value for
block. The input variables to the block are:
T 1H , however, we first need to account for the Peltier effect which
acts on the second term on the right of Eq. (8). The Peltier effect
 TEG geometrical parameters: surface A and thickness L.
pumps additional power from the hot side to the cold side, thus
 Conduction and convection coefficients: open-circuit thermal
effectively reducing the hot side temperature T H provided by the
conductivity k and thermal diffusivity coefficient
HE. If we write that Pconv H ¼ hH ADT H ¼ hH AðT 1H  T H Þ, where T H
thermal conductiv ity k
 ¼ v olumetric heat capacity qC p
of the TEG; thermal transfer coeffi- is the result of the HE, then we can correct the value of T H by cal-
cients through the hot and cold mediums bH and bC . culating its ‘real’ value:
170 A. Montecucco, A.R. Knox / Applied Energy 118 (2014) 166–172

Pconv H þ PPeltH L a2 T H þ T C
T Hreal ¼ T 1H  ð9Þ ZT ¼ ð16Þ
hH A Rint A k 2
where PPeltH ¼ aIT H from Eq. (2). Now that the ‘real’ value for T H is V load Iload
g¼ ð17Þ
obtained, from Eq. (8) it is possible to calculate T 1H for the next Pin  PlossH
iteration: In Eq. 16 the electrical conductivity was written as r ¼ L=ðRint AÞ and
  the average temperature of the device as T AVG ¼ ðT H þ T C Þ=2.
Pin  PlossH  hH AðT 1HðiÞ  T Hreal Þ tstep
T 1Hðiþ1Þ ¼ T 1HðiÞ þ ð10Þ The electrical part of the Simulink model computes the values
mH C H
of Rint and V load depending on the actual temperature gradient DT
It can be noted that, depending on the polarity of the second term and on the electrical parameters Iload ; N S ; N P . The Peltier term is cal-
on the right of Eq. 10, T 1Hðiþ1Þ can be greater or smaller than culated as N tot aT j Isingle , where Isingle is the current passing through a
T 1HðiÞ . T Hreal and T 1Hðiþ1Þ will be used in the next iteration ði þ 1Þ of single TEM. The load current (or voltage) in a real TE system is set
the HE as the initial value of the hot side temperature and as the either by a constant Rload connected to the TEG, or by an interfacing
temperature of the hot thermal mass, respectively. power electronic converter with Maximum Power Point Tracking
For the cold part of the system similar considerations hold true. (MPPT) [29]. For flexibility of simulation, the desired load is passed
The power flowing to the cold thermal mass is the sum of the ther- to the model as an input, but the model provides also the maxi-
mal power flowing through the TEG and the thermal power mum theoretical power, i.e. when V load ¼ V OC =2, to compare actual
pumped to the cold junction by the Peltier effect. In a similar electrical power output with the maximum that can be extracted.
way as for the hot side we update T C to its ‘real’ value:
hC AðT C  T 1C Þ þ aIT C 4. Experimental and simulation results
T Creal ¼ T 1C þ ð11Þ
hC A
In order to test the ability of the proposed Simulink–Matlab
It is now possible to obtain T 1Cðiþ1Þ as model to simulate real TE generating systems, a transient experi-
 
hC AðT Creal  T 1CðiÞ Þ  Pout  PlossC tstep ment has been performed in the laboratory and then simulated
T 1Cðiþ1Þ ¼ T 1CðiÞ þ ð12Þ by computer. The results for both and a discussion about their
mC C C
comparison is presented in this Section.
where Pout is the power removed from the cold thermal mass and
PC out takes into account any thermal power lost to ambient on the 4.1. Experimental test rig
cold side.
Both the values calculated in Eq. (11) and (12) are passed to the The TE system used in the experiment is identical to that pre-
HE block for the next iteration ði þ 1Þ. sented in [28]. The cold thermal mass is formed of a water-cooled
copper block and the hot thermal mass is a copper block containing
3.4. The electrical block one high-power high-temperature electrical heater powered by a
DC power supply unit. The TEG output is connected to an electronic
A TEG can be modelled as a voltage source V OC in series with its load. The TEG module is compressed between the two blocks by a
internal resistance Rint , even during transients, because their elec- mechanical fixture which does not introduce thermal shocks or
trical dynamic response is in order of nanoseconds [26]. Therefore, short-circuit the thermal path. The mechanical compression force
when the TEG is connected to a load Rload it can be written that the on the system is sensed by a load cell and thermocouples measure
load voltage is the temperatures of the water, the heater, and the faces of the TEG
V load ¼ V OC  Rint Iload ð13Þ module (directly touching them). All the sensors are connected to a
datalogger. Each of the electronic instruments used is connected by
It can be noted that both the internal resistance and the open-cir- the IEEE-488 interface (GPIB) to a computer and controlled by a
cuit voltage can be approximated to linear functions of the temper- program created in the Agilent VEE Pro software.
ature difference DT across the device [27], therefore Eq. (13) can be The copper blocks have a thermal capacity of approximately
updated to 565 J/K (Copper volumetric heat capacity C V Cu ¼ 3:45 J=cm3 K),
V load ¼ mV OC DT  ðmRint DT þ qRint ÞIload ð14Þ the thermal transfer coefficient for between the TEG and the hot
and cold thermal masses are respectively 9100 and 11,800 W/
As stated in the Introduction, this work neglects the Thomson ef- m2 K (Copper thermal conductivity kCu ¼ 400 W=mK; the heater
fect; therefore, mV OC is considered constant and equal to the Seebeck is 2.2 cm from the TEG hot side, the water pipes are 1.7 cm from
coefficient a (V/K). mRint (X/K) and qRint (X) are constant coefficients, the TEG cold side).
too. The input power is provided by electrical heaters and thermal
In order to use Eq. 14 in the model, the appropriate values for energy is removed from the cold side through forced liquid flow.
the coefficients mV OC ; mRint and qRint should be obtained through an The mass flow rate of the water/glycol mixture is measured as
electrical characterisation of the TEGs used in TE system [28]. 17 g/s and the coolant mixture heat capacity is 3.89 J/gK. The con-
The electrical block also deals with series and parallel connec- vection coefficient was determined empirically and is 1400 W/
tion of TEG modules. It is assumed that all the modules are identi- m2 K. Heat losses from the hot block (which is insulated with
cal and all are subjected to equal thermal conditions. If there are N S glass-fibre) are included in the simulation as described in [28].
modules in series and N P modules in parallel, for a total of N tot , the The time step size used for the simulation is 3 s. This choice is
total open-circuit voltage is V OCtot ¼ N S V OC and the load current Iload based on the fact that the test rig is able to vary the temperature
is N P times the current in a single branch. It can then be written, difference at a maximum rate of 0.2 K/s, therefore a time step of
from Eq. (13), that 3 s ensures that the temperatures of the thermal masses do not
change much and at the same time guarantees a good evolution
NS Rint Iload
V load ¼ NS V OC  ð15Þ of the transient solution provided by the heat equation.
NP
The TEG module used is manufactured by European
The dimension-less figure of merit ZT and the actual efficiency g are Thermodynamics Ltd. (model code: GM250-127-14-10) and has
also calculated [14,9]: an open-circuit thermal conductivity k ¼ 1:62 W=mK, surface
 A. Montecucco, A.R. Knox / Applied Energy 118 (2014) 166–172 171

A ¼ 1600 mm2 , thickness L ¼ 3 mm, thermal diffusivity This experiment is designed to test the simulation model under
 ¼ 1e6 m2 =s. thermal and electrical transients of difference magnitude. Different
Electrical characterisation of the module used was performed at temperature variation rates were applied both in open-circuit and
two different temperature differences, DT ¼ 50  C and at-load, with several electrical load steps. The Simulink program
DT ¼ 150  C, both with a mechanical force of 2000 N on the mod- runs without interruptions, therefore testing it under different
ule, corresponding to a pressure of 1.2 MPa. Results obtained are thermal and electrical conditions ensures that there are no offsets
shown is Fig. 4. These data are then used to obtain the coefficients that would make the results valid for just one of those operating
of Eq. (14): mV OC ¼ 0:048; mRint ¼ 0:0045; qRint ¼ 1:52. conditions.
Fig. 5 shows both the experimental and simulation results in
4.2. Comparison and discussion of results the same graph, for comparison. In particular the temperature dif-
ference and the output electrical power are plotted versus the tran-
The designed transient lasts for 55.5 min and is intended to pro- sient time. During the transient the temperature difference across
duce several power and load steps. The VEE program records all the the TEG device varies from 5 °C to almost 150 °C, leading to some
electrical data, temperatures, mechanical pressure and time. The thermal expansion effects in the whole system. The mechanical
transient parameters are as follows, where Pin is the DC electrical pressure on the system could not be controlled during the test,
power to the heater and Iload is the current value set in the elec- but active control of the clamping force will be incorporated into
tronic load: a future test rig to compensate for thermo-mechanical effects. In
previous work [28] it was noted that different pressure loads
1. P in ¼ 50 W;Iload ¼ 0 A (open-circuit) @ time = 0 s. slightly influence the contact resistances in the system, thus affect-
2. P in ¼ 50 W;Iload ¼ 0:2; 0:4; 0:6; 0:8; 1 A @ time = 1195, 1210, ing the thermal conductivity and the power produced by the TEG
1229, 1247, 1286 s. module. Their effect should not account for more than a 5% change
3. P in ¼ 150 W;Iload ¼ 1 A @ time = 1286 s. in the results.
4. P in ¼ 150W; Iload ¼ 1:2; 1:4; 1:6; 1:4; 1:2; 1:0; 0:8; 0:6 A @ time = As it can be appreciated from Fig. 5, there is no delay in the re-
2761, 2780, 2795, 2828, 2844, 2877, 2946, 3109 s. sponse of the simulated results with respect to the experimental
data. The simulation results show very good agreement with the
experimental results, accurately tracking the electro-thermal cou-
pled effects occurring in the TE system. The Root Mean Square Er-
ror (RMSE) is 3.58 °C for the temperature difference and 0.2 W for
the power output. The RMSEs normalised to the values range
(NRMSE) become 2.76% and 4.55% for the temperature difference
and the output power, respectively. There are no works in the lit-
erature that proposed a simulation model for TEG systems demon-
strating similar prediction performance under varying thermal and
electrical conditions as the one presented in this article. As a con-
sequence it is difficult to compare the data presented here to other
simulation results reported in the literature.
It is very interesting to note the close effect that the load cur-
rent has on the temperature gradient across the device. Every in-
crease in Iload leads to a decrease in temperature due to the
greater value of the Peltier term of Eq. (9). Conversely, the last
500 s of the test show that a smaller load current increases the rate
of change of temperature. The effective thermal conductivity of the
TEG module varies markedly, depending on the current flowing
through the module. MPPT converters should take this thermal
Fig. 4. Electrical characteristic of the TEG module used in the experiment, for two
different temperature gradients DT = 50, 150 °C, clamped at 2 kN/1.2 MPa.

Fig. 5. Experimental and simulated transient results showing the temperature difference and the output electrical power versus the time.
172 A. Montecucco, A.R. Knox / Applied Energy 118 (2014) 166–172

variation into account when setting the optimum operating point [8] Risse S, Zellbeck H. Close-coupled exhaust gas energy recovery in a gasoline
engine. Res Therm Manage 2013;74:54–61.
for maximum efficiency or power generation.
[9] Wang Y, Dai C, Wang S. Theoretical analysis of a thermoelectric generator
Future work will develop a model of the system that incorpo- using exhaust gas of vehicles as heat source. Appl Energy 2013;112:
rates the non-linear effects introduces by a MPPT converter. The 1171–80.
model will be adapted to a typical waste heat energy recovery sys- [10] Patyk A. Thermoelectric generators for efficiency improvement of power
generation by motor generators environmental and economic perspectives.
tem from exhaust gas for automotive applications. Appl Energy 2013;102:1448–57.
[11] Champier D, Bédécarrats J, Kousksou T, Rivaletto M, Strub F, Pignolet P. Study
of a TE (thermoelectric) generator incorporated in a multifunction wood stove.
5. Conclusion Energy 2011;36:1518–26.
[12] OShaughnessy S, Deasy M, Kinsella C, Doyle J, Robinson A. Small scale
This paper presents an innovative and powerful computer tool electricity generation from a portable biomass cookstove: prototype design
to accurately simulate real thermoelectric power generating sys- and preliminary results. Appl Energy 2013;102:374–85.
[13] Suter C, Jovanovic Z, Steinfeld A. A 1 kWe thermoelectric stack for geothermal
tems even during transient conditions. The proposed simulation
power generation modeling and geometrical optimization. Appl Energy
program, designed in Simulink–Matlab, deals with all the most 2012;99:379–85.
important thermoelectric effects and is able to cope with both [14] Sark WV. Feasibility of photovoltaic thermoelectric hybrid modules. Appl
thermal and electrical dynamics. Consequently it can greatly help Energy 2011;88:2785–90.
[15] Xiao J, Yang T, Li P, Zhai P, Zhang Q. Thermal design and management for
the design phase of large-scale and/or complicated thermoelectric performance optimization of solar thermoelectric generator. Appl Energy
systems. 2012;93:33–8.
A comparison of experimental and simulation results shows the [16] Qiu K, Hayden aCS. Development of a novel cascading TPV and TE power
generation system. Appl Energy 2012;91:304–8.
accuracy and capability of the model, showing that it can be em- [17] Biswas K, He J, Blum ID, Wu C-I, Hogan TP, Seidman DN, et al. High-
ployed to study transients and steady-state operation of real TE performance bulk thermoelectrics with all-scale hierarchical architectures.
systems with confidence. Nature 2012;489:414–8.
[18] Zebarjadi M, Esfarjani K, Dresselhaus MS, Ren ZF, Chen G. Perspectives on
thermoelectrics: from fundamentals to device applications. Energy Environ Sci
Acknowledgements 2012;5:5147.
[19] Lineykin S, Ben-Yaakov S. Modeling and analysis of thermoelectric modules.
The authors would like to thank Dr. Steven Roper and Dr. James IEEE Trans Indust Appl 2007;43:505–12.
[20] Chen M, Rosendahl LA, Condra TJ, Pedersen JK. Numerical modeling of
Buckle of the University of Glasgow for their contribution to the
thermoelectric generators with varing material properties in a circuit
project and Mr. Peter Miller and the mechanical workshop of the simulator. IEEE Trans Energy Convers 2009;24:112–24.
University of Glasgow for their work in creating the test rig used [21] Naji M, Alata M, Al-Nimr MA. Transient behaviour of a thermoelectric device.
Proc Inst Mech Eng Part A: J Power Energy 2003;217:615–21.
for the experiments.
[22] Cheng C-H, Huang S-Y. Development of a non-uniform-current model for
predicting transient thermal behavior of thermoelectric coolers. Appl Energy
References 2012;100:326–35.
[23] Meng J-H, Wang X-D, Zhang X-X. Transient modeling and dynamic
[1] Du C-Y, Wen C-D. Experimental investigation and numerical analysis for one- characteristics of thermoelectric cooler. Appl Energy 2013;108:340–8.
stage thermoelectric cooler considering Thomson effect. Int J Heat Mass [24] Montecucco A, Buckle JR, Knox AR. Solution to the 1-D unsteady heat
Transfer 2011;54:4875–84. conduction equation with internal Joule heat generation for thermoelectric
[2] Lazard M. Heat transfer in thermoelectricity: modelling, optimization and devices. Appl Therm Eng 2012;35:177–84.
design. In: 7th IASME/WSEAS international international conference on heat [25] Lee H. Optimal design of thermoelectric devices with dimensional analysis.
transfer, thermal engineering and environment. p. 129–34. Appl Energy 2013;106:79–88.
[3] Chen J, Yan Z, Wu L. The influence of Thomson effect on the maximum power [26] Chen L, Cao D, Yi H, Peng FZ. Modeling and power conditioning for
output and maximum efficiency of a thermoelectric generator. J Appl Phys thermoelectric generation. In: IEEE power electronics specialists
1996;79:8823. conference. IEEE; 2008. p. 1098–103.
[4] Min G, Rowe DM, Kontostavlakis K. Thermoelectric figure-of-merit under large [27] Woo BC, Lee DY, Lee HW, K IJ. Characteristic of maximum power with
temperature differences. J Phys D: Appl Phys 2004;37:1301–4. temperature difference for thermoelectric generator. In: 20th International
[5] Riffat S, Ma X. Thermoelectrics: a review of present and potential applications. conference on thermoelectrics; 2001. p. 431–4.
Appl Therm Eng 2003;23:913–35. [28] Montecucco A, Buckle J, Siviter J, Knox AR. A new test rig for accurate
[6] Siviter J, Knox A, Buckle J, Montecucco A, McCulloch E. Megawatt scale energy nonparametric measurement and characterization of thermoelectric
recovery in the Rannkine cycle. In: Energy conversion congress and exposition generators. J Electron Mater 2013;42(7):1966–73.
(ECCE). IEEE; 2012. p. 1374–9. [29] Montecucco A, Siviter J, Knox A. Simple, fast and accurate maximum power
[7] Crane D, LaGrandeur J, Ranalli M, Adldinger M, Poliquin E, Dean J, et al. TEG on- point tracking converter for thermoelectric generators. In: Energy conversion
vehicle performance and model validation and what it means for further TEG congress and exposition (ECCE). IEEE; 2012. p. 2777–83.
development. In: International conference on thermoelectrics (ICT’12).