Real-Time Monitoring of Thermal Response andLife-Time Varying Parameters in Power Modules

Christoph H. van der Broeck∗,∗∗, Timothy A. Polom∗∗, Robert D. Lorenz∗∗ and Rik W. De Doncker∗∗Institute for Power Electronics and Electrical Drives (ISEA), RWTH Aachen University

Jaegerstrasse 17-19, 52066 Aachen, Germany, Email: post@isea.rwth-aachen.de∗∗ Wisconsin Electric Machines and Power Electronics Consortium (WEMPEC), University of Wisconsin-Madison

1415 Engineering Drive, 53706 Madison, WI, Email: admin@wempec.wisc.edu

Abstract—This paper introduces new technologies for mon-itoring thermal response and life-time varying parameters ofpower electronic modules in real-time without interruptingnormal converter operation. The technologies are embeddedin an adaptive thermal observer that integrates temperaturemeasurements and electrothermal models to estimates devicelosses and temperatures at reliability-critical locations of thepower module. The observer structure includes a model ref-erence adaptive system (MRAS) for adaptively calibrating de-vice switching-loss models. This improves thermal monitoringaccuracy, avoids time-consuming off-line calibrations and createsthe opportunity for efficiency estimations during power moduleoperation. The adaptive observer together with small-signalloss-excitation and system-identification technologies extract thethermal impedance of the power module in magnitude andphase over multiple frequency decades. This in-situ thermalimpedance spectroscopy (ITIS) has the unique feature of con-tinuously providing information on the thermal interface of thepower devices without interrupting normal inverter operation.The extracted information, i.e. life-time varying electrical andthermal parameters, enable diagnosing power module state-of-health with improved accuracy compared to prior technologies. Iteven allows separating different degradation effects, such as bondwire lift-off, device and base-plate solder delamination as well asdeterioration of the convection process. With these new features,next generation monitoring systems can extract thermal responseand degradation and thus protect power modules from severethermal stress and overload. Furthermore, the obtained state-of-health information can be selectively used to reduce thermalstress and to schedule predictive maintenance leading to a saferand more reliable operation of future converter systems.

I. INTRODUCTION

The power density of power electronic systems has in-creased significantly over the last decades due to numerousimprovements in the multi-physics integration of power de-vices [1], [2]. Continuing this trend is of great importance forthe development of next generation electric vehicles [3], [4].However, an even stronger miniaturization of power electronicmodules without jeopardizing reliability and safety is onlypossible with a simultaneous development of electrothermaland degradation monitoring systems [5]–[12]. They guaranteereliable and safe converter operation by real-time monitoringand control of thermal stress [13]–[21], early diagnosis ofcritical degradation [9], [11], [22], [23] and timely initiationof predictive maintenance [24]–[27].

For the real-time electrothermal monitoring of power elec-tronic modules, observers are effective tools that can esti-mate temperatures and device losses [15], [18], [28]–[30].Observers, which have been used in power converters and

drives for decades [31]–[33], were first proposed for elec-trothermal monitoring of power electronic devices in [8]. Inthis application, they combine junction temperature informa-tion with electrothermal real-time models to estimate losses,junction temperature, and 3-D distributed temperatures, e.g.at critical module interconnects, with minimal sensitivity tomodel inaccuracies [18]. The junction temperature informa-tion is either obtained on the basis of temperature-sensitiveelectrical or optical parameters [34], [35], or it is measuredwith on-chip temperature sensors [36]. Depending on theapplication, a wide range of thermal real-time models canbe applied in observers. For estimating junction temperaturesand device losses only, a 1-D Cauer model [17], [37] mayprovide sufficient information and accuracy. However, if 3-D distributed temperatures shall be estimated in real time,the combination of finite difference modeling [38] and modeltruncation techniques [39]–[41] yield compact and precisereal-time models. The required model derivation process isdiscussed in [30], [42], [43].

This work presents an adaptive thermal observer derivedfrom [18] that monitors instantaneous, i.e. switching periodaveraged, and excitation period averaged device losses andtemperatures with nearly zero lag and minimal sensitivityto model inaccuracies . The instantaneous device losses andtemperature estimation is essential for a high-bandwidth elec-trothermal monitoring system that is required for precisedata acquisition and fast thermal overload protection [8],[15], [28], [44]. The excitation period averaged device lossesand temperature are key control variables for active thermalcontrol of power electronic inverters and must be free of lag[14], [21], [45], [46].

Inaccuracies of device loss models are a strong limitation ofall loss and temperature estimation methods. Unlike thermalmodels of fluid cooled power modules, which can be derivedwith a comparatively high accuracy of approximately 5 %[47], [48], the extraction of accurate switching loss modelsis difficult: The precise measurement of switching transientsand consequent extraction of switching loss energies overthe entire operation range of a power module requires ahigh expertise on modeling measurement equipment [49].However, even cautious state-of-the-art modeling results ina best case accuracy of 20 % [50]–[52]. Furthermore, thedevice characteristics and parasitics of different converters thatare realized with the same type of power module vary [53].This means that a generic loss model must be calibrated for

every power converter individually. The resulting process istime consuming and hence not feasible for a fast majority ofapplications. Thermal observers can intrinsically compensateinaccuracies of device loss models within their feedback band-width. However, above the observer feedback-bandwidth, anydevice-loss inaccuracies results in deviations of the tempera-tures estimates [18]. To overcome this limitation, a adaptivedevice loss model is introduced in this work that utilizes amodel reference adaptive system for self-calibration.

Another important scope of this work lies in monitoringthe thermal interface characteristics of the power moduleand consequently diagnosing degradation mechanisms. State-of-the-art approaches have identified the static junction-to-ambient thermal resistance as an indicator for delaminationprocesses [5] that can be extracted online according to [54],[55]. However, their accuracy is bounded by the previouslyaddressed poor precision of most device loss models. Also,these techniques are unable to distinguish different degrada-tion mechanisms. The transient thermal impedance Zth pro-vides more in-depth information about localized delaminations[56]–[59]. It is estimated either by external excitation with anadditional circuitry [41] or by generating a load current stepvia the converter control algorithm according to [58], whenthe normal converter operation interrupted. However, none ofthese approaches can extract the transient thermal impedancewithout interrupting the normal converter operation.

To address these limitation, this work proposes a newmethod for extracting the thermal impedance of the powermodule in the frequency domain referred to as in-situ thermalimpedance spectroscopy. In-situ thermal impedance spec-troscopy (ITIS) identifies the magnitude and phase of thepower module thermal impedances over multiple frequencydecades without interrupting normal converter operation. Itapplies periodic small-signal device-losses and subsequentlyextracts the excitation-response via temperature sensing andfiltering to reconstruct the thermal impedance of the powermodule. As the thermal impedance can be extracted withnegligible error, it allows separate identification of differentdegradation mechanisms [60], e.g. die-attach or DCB-attachsolder layer due to fatigue [61] as well as a deteriorationof the convection process [62]. The extracted state-of-healthinformation allows operating the converter at reduced stresslevels if a critical degradation is reached , e.g. by reducing therated power or by active thermal control [7], [8], [13], [14],[63]. Furthermore, the degradation information can be used toschedule a predictive-maintenance of power converters [24]–[26] to ensure in-time replacement of critical components.

This paper, which is based on the work in [64] and [65],is organized as follows: Section II reviews the electrothermalmodels and applicable junction temperature measurement con-cepts that are the basis for the thermal and state-of-healthmonitoring structure. In Section III, the thermal observerstructure is presented that estimates losses and 3-D distributedtemperatures. Section IV shows how the thermal monitoringsystem can be enhanced with an adaptive switching losscalibration yielding more precise thermal monitoring. SectionV introduces in-situ thermal impedance spectroscopy (ITIS)

Fig. 1. Hybridpack2 (HP2) power module (left) and meshing of its finitedifference model (right)

and illustrates how it can be used to track life-time varyingthermal parameters and diagnose degradation in power mod-ules. Finally, in Section VI and Section VII the performanceof the monitoring system and the self-sensing algorithms areevaluated in simulations and experiments for the Hybridpack2(HP2) power module (FS800R07A2E3B13) depicted in Fig.1.

II. ELECTROTHERMAL MODELING AND JUNCTIONTEMPERATURE MEASUREMENT

This section introduces the electrothermal model and thejunction temperature measurement that are used for this work.

A. Thermal Model

A key component of the thermal monitoring system is thethermal real-time model of the power module. This workuses 3-d finite difference modeling (FDM) to subdivide apower module into small cubes, as illustrated in Fig. 1. Thethermal characteristic of each cube is represented by a thermalcapacitance and thermal resistances to its neighbors, resultingin a large thermal equivalent circuit represented by Fig. 2. Theentire thermal equivalent circuit of the power module can besummarized in a state space model (1).

dθ

dt= A · θ +B · P loss T out = C · θ (1)

The model (1) requires the device losses P loss, the transitionmatrix A, the input matrix B and the output matrix C tocalculate the temperature vector T out. The output temperaturevector T out includes temperatures at all four devices, tem-perature nodes below the devices at the device solder layer,at the DCB-to-baseplate solder layer, at the NTC and thespatially-averaged base plate temperature. These local nodesallow monitoring temperatures of critical material interfacesand are visualized in Fig. 3.

With model reduction techniques, in this case the balancedtruncation method, the state space model is significantlyreduced in size such that is applicable as a real-time model.The final model is characterized by the matrices A, B andC. For the HP2 power module in Fig. 1, the final model has14 states and can estimate the thermal behavior accuratelywithin a bandwidth of 100 Hz. The model is discretized withthe zero-order hold method [66], [67] with a sampling timeof Ts = 1 ms to make it applicable on a digital signalprocessor. The sampling time of Ts = 1 ms achieves anaccurate discretization if the losses do not vary significantlyover one sampling time, as discussed in [42]. A more detailed

p

Dz

bRth,d

bRth,l

qloss

aRth,r

aRth,baRth,u

aRth,d

aRth,f

aRth,l

Cth

bRth,r

bRth,u

m,n,pT

x

y

z

m-1,n,pT m+1,n,pT

m,n+1,pTm-1,n+1,pTm+1,n+1,pT

m,n-1,pTm-1,n-1,pT m+1,n-1,pT

mDx m+1Dxm-1Dx

n-1

Dy

nDy

n+1

Dy

Fig. 2. Excerpt thermal network resulting from application of FDM

Device Area IGBTA DiodeA DiodeB IGBTB

Device

Upper Sd Layer

Lower Sd Layer

NTC Base Plate

Si

Sd

Cu

Al O2 3

Cu

Sd

Cu

H O2

AC

DC+

DC-

Fig. 3. Monitoring nodes of the extracted thermal real-time model depictedfor one phase of the power module

derivation of the electrothermal model can be found in [29],[41], [42] and [18].

B. Loss Model

Next, a model for the prediction of instantaneous andaveraged losses of the IGBT and the diode is presented basedon [42].

1) Instantaneous loss model: First, the conduction lossesof the IGBT and the diode are addressed. They depend onforward voltages and currents as shown in (2).

P IGBTcond = vIGBT

ce (ice,Tj) · ice and PDiodecond = vDiode

T (id,Tj) · id(2)

The forward voltages vIGBTce and vDiode

T are modeled based onextracted data sheet parameters from [68], summarized in Tab.I. They are approximated as a function of the device junctiontemperature Tj and the device currents ice and id via (3).

vIGBTce (ice,Tj) = Vce(Tj) +Rce(Tj)ice + Sce(Tj)

√ice

vDiodeT (id,Tj) = VT(Tj) +RT(Tj)id + ST(Tj)

√id (3)

TABLE ICONDUCTION LOSS MODEL PARAMETERS OF THE HP2

Tj Vce Rce Sce VT RT ST

25 °C 0.54 V 0 Ω 0.03 V/A0.5 334 mV 0 Ω 0.06 V/A0.5

125 °C 0.31 V 0.2 mΩ 0.04 V/A0.5 222 mV 0 Ω 0.06 V/A0.5

TABLE IISWITCHING LOSS MODEL PARAMETERS OF THE HP2

E0 K0 α β KT Err0 Krec

0 KrecT

1.2 mJ 0.1 mJ/A 1.75 0.82 1 µJ K−10.5 mJ 4.4 µJ A−1 0.02 K−1

∗for V refdc = 400 V , Rref

g = 2.2 Ω and T ref = 20 °C

The switching losses of the IGBT and the diode are predictedwith parametric loss models given by (4) and (5) as a functionof the device currents ice and id, the dc-link voltage Vdc, thegate resistance Rg and the device temperatures differences∆Tj with respect to the reference temperature T ref

j . Theparameters of the model are summarized in Tab. II and werederived in [42].

P IGBTsw =

(E0 +K0ice

(Vdc

V refdc

)α(Rg

Rrefg

)β+ ∆TjKT

)· fsw

with ∆Tj = Tj − T refj (4)

PDsw =Erec

0 +Krec0 ice

(Vdc

V refdc

)α(Rg

Rrefg

)−β

(1 + ∆Tj ·KrecT ) · fsw

with Erec0 = Err

0 ·Vdc

V refdc

(5)

2) Averaged loss model: Based on the instantaneous lossmodel, a model for the averaged device losses is derived byintegration of (2), (4) and (5) over one excitation period T0of the current resulting in (6) - (9).

P IGBTcond = VceI

(1

2π+M cos(ϕ0)

8

)+RceI

2

(1

8+M cos(ϕ0)

3π

)+ Sce · I

32 · (0.139 + 0.1144 ·M cos(ϕ0)) (6)

PDcond = VTI ·

(1

2π− M cos(ϕ0)

8

)+RTI ·

(1

8− M cos(ϕ0)

3π

)+ ST · I · (0.139− 0.1144 ·M · cos(ϕ0)) (7)

P IGBTsw =

(E0

2+K0

πI

(Vdc

V refdc

)α(Rg

Rg

)β+

∆TjKT

2

)· fpwm (8)

PDsw =

Erec0

2+Krec

0

πI

(Vdc

V refdc

)α(Rg

Rrefg

)−β

· (1 + ∆TjKrecT ) · fpwm

(9)

The averaged losses can be computed based on the modulationindex M = V /(Vdc/2), the current amplitude I , the loadangle ϕ0, the gate resistance Rg, the device temperatures Tjand the parameters in Tab. I and Tab. II.

C. Junction Temperature Measurement

The monitoring technologies that are developed in this workrequire to measure junction temperature. Various junction tem-perature extraction methods have been proposed in literature:Temperature sensitive electrical parameters (TSEPs) can bemeasured, e.g. with smart gate drivers, to extract junctiontemperature information [34], [69]. Examples of TSEPs aredevice forward voltage [70]–[73], gate current [74], [75],turn-on delay [76], [77], gate resistance [78] or gate voltage

Power Device

Power Module

Measurement Delay

Measurement Noise

Thermal Interface

Loss Estimation

Adaptive Loss Model Identification

Thermal Observer

Section IV

Section V

Section III.

Converter

Operation

Ki

K (I)i

Tspatial

T (k)spatial

T (k)spatial

T (k)spatial

T

T

T

Tj

T (k)j

T (k)j

T (k)j

T (k)j

T (k-n)j

T (k-n)j

T (k-n)j

DTj

DTj

Ploss

DPloss

P (X ,DR )loss conv gate

P (X )loss conv

P (X )loss conv

Ploss

Ploss

Ploss

Ploss

Ploss,S

Ploss,S

X conv

avg

avg

avg

avg

avg

avgavgavg avg

avg avg

t

t

t

t t

t

t

avg

ttt

tt

avgavg

avg

DP (k)loss

DP (k)loss

DPloss

DPloss

DPloss

-1T zs

-1T zs

Vdc

fsw

Dfsw

Dfsw

Rgate

Tj

abcd

DK0

dDK0

dt

DK0

abci

0

Kp

dT

-nz

-nz

-11 - z

-11 - z

k

= AT + B P loss

T (k)= C q (k)

T (k)= C q (k) + Tf

q (k+1)= A q (k)+B P (k) loss

q (k+1)= A q (k)+B P (k) loss

dt

out

out

out

out

d

d

d

d

d

d

Low-Frequency Temperature Estimation

High-Frequency Temperature Estimation

DRgate

DRgate

A

In-situ Thermal Impedance Spectroscopy (ITIS)

Orthogonal Correlation

Tspatial

TjZ (jw )th 0

T

DTj

ex

ex exexex ex

exex

T (k)= C q (k)

q (k+1)= A q (k)+B P (k) loss

out

outd d

d

Excitation-Frequency Temperature Estimation

Kr

-1T (1- cos(w T )z )s 0 s

-1 -11-2cos(w T )z + z0 s

P0P0

Ploss

Ploss

DPloss

DRgate

w0

sin(w t)0

DR (X )g conv

ex

ex

ex

B

B

Fig. 4. Thermal and state-of-health monitoring system for power electronic modules comprising a thermal observer structure (Section III), an adaptive lossmodel calibration unit (Section IV) and a thermal impedance spectroscope (Section V)

[79]. Recent research also investigated the electrolumines-cence of the body diode in SiC devices as a temperaturesensitive optical parameter [80]–[83] Alternatively, on-chipjunction-temperature sensors [36], [84], [85] can be applied.This work utilized a decapsulated power module and high-bandwidth infrared (IR) sensors, which were placed abovethe two diodes and IGBTs of a half bridge, for experimen-tal evaluation. Thereby the complexity of the experimentalsetup was reduced, without any limitation for evaluation andtesting. However, in an integrated converter, the IR sensorsmust be replaced by one the discussed temperature extractionapproaches.

III. THERMAL OBSERVER STRUCTURE

The electrothermal real-time model and the junction tem-perature measurement are combined in a thermal observerstructure depicted in Fig. 4. This section describes the op-eration of the observer that estimates 3-D distributed temper-atures as well as device losses with minimal error. Conse-quently, it is explained how the observer can detect modelerrors that can either be used for calibration or trackingelectrothermal parameters that change due to degradation.

A. Functional Principle

During basic operation of the observer, the adaptive lossmodel identification (red) and the excitation temperature esti-mation (blue) can be neglected, i.e. the indicated switchesA and B in Fig. 4 are open. First, the thermal observercomputes instantaneous and averaged losses P loss and P

avg

loss

based on the operation vector Xconv. In a next step, theaverage losses are subtracted from the instantaneous lossesto determine the transient losses P

t

loss, which exhibit onlyhigh bandwidth loss information. The transient losses andthe averaged losses are passed to two identical instances ofthe reduced-order thermal model to compute the transientand the averaged temperatures within the power module. Thetemperature estimates are summarized in the vectors T

t

out andT

avg

out and its locations are illustrated in Fig. 3.The transient temperature vector T

t

out only exhibits the highbandwidth temperature information, whereas the averagedtemperature vector T

avg

out exhibits only low bandwidth temper-ature information. If both temperature vectors are summed, theinstantaneous temperatures T out are obtained. The junctiontemperature estimates T j, which are embedded in T out, aredelayed by the measurement delay n to ensure consistency and

are compared with the measured junction temperatures T j. Ifthere is an error between the junction temperature predictionand the measurement, this error ∆T j is multiplied by aproportional feedback gain Kp to correct the transient temper-ature estimation. The error is also passed to an integrator withthe gain Ki such that the accumulated error can correct theaverage loss prediction. Thereby, the high bandwidth error isused to correct the transient temperature estimation, whereasany error at low bandwidth including dc is used to correctthe average temperature estimation. This technique, referredto as bandwidth partitioning, has been developed based on theconceptual background of [32], [86].

The maximum bandwidth at which the proportional statefeedback path can correct the temperature and loss estimatesof the model can be adjusted with the feedback gain Kp.By adjusting the ratio between integral and proportionalfeedback gain Kp/Ki, the bandwidth at which the separationof averaged and transient estimates occurs can be changed.Fig. 5 shows that the observer can estimate instantaneous andaverage junction temperature within the integral bandwidthwithout error even if the loss model has errors. Above theintegral bandwidth, model errors lead to small magnitudeestimation errors that moderately grow to the bandwidth fo theproportional feedback path. Above the proportional feedbackpath the magnitude accuracy only depends on the accuracyof the electrothermal model as the observer cannot correctestimates above its feedback bandwidth. Most importantly,over the entire bandwidth the observer estimates temperatureswith nearly zero lag in phase independent of modeling errors.This zero phase-lag property is critical for active thermalmanagement and control [14]. A more detailed analysis anddiscussion of the observer, in particular of its feedback designis provided in [29] and [18].

B. Utilizing Correction Variables for Calibration and Degra-dation Detection

The correction variables, ∆Pavg

loss and ∆Pt

loss of the ob-server support three major tasks:

Firstly, they correct the temperature estimates within thefeedback bandwidth of the observer. This basic task wasdescribed in the previous paragraph.

Secondly, the correction variables allow closed loop lossestimation for power modules whose thermal interface is accu-rately modeled. This is the case for liquid cooled power mod-ules, whose structure and convection interface can be modeledwith good accuracy if it has not experienced degradation.Thus, transient and averaged feedback corrections ∆P

avg

loss

and ∆Pt

loss are almost exclusively caused by inaccuraciesof the device loss model. The closed loop correction of thedevice loss model enables highly accurate loss estimation ofindividual power devices, i.e. IGBTs and diodes Consequently,it can even detect hard- and soft-switched operation of powerdevices.

Finally, the the correction variables can be used to calibratethe device loss model that is used for loss estimation. A cali-bration is desirable as it enables accurate estimation of devicelosses and thus temperatures even above the observer band-width. Thus, in the following section, a method is introduced

0

0.5

1

1.5

2

10-2 10-1 10

Frequency in Hz

0 101 102

-90

-45

0

45

90

-20% Loss Estimation Error+20% Loss Estimation Error

Perfect Loss Estimation

Proprtional bandwidth

Ob

se

rve

rD

esi

gn

Integral bandwidth

Resonator frequency f = 10 Hzex

f = 1Hzb,i

f = 10 Hzb,p

Excitation TemperatureAveraged Temperature

Instantaneous Temperature0.5

1

8 9 10 11 12

8 9 10 11 12

-90-45

04590

Fig. 5. Observer estimation accuracy showing that independent of potentialmodeling errors: A) Instantaneous and averaged junction temperatures areestimated over wide bandwidth with minimized magnitude error and zerophase lag B) Temperature response to loss excitation is separated fromremaining temperature profile and estimated precisely in magnitude and phase

Fig. 6. Observer-based model calibration and degradation detection

to calibrate a parametric loss model and drive the correctionvariables to zero. In Fig. 6 it is illustrated how this calibrationleads to a strong reduction of the correction variables ∆P

avg

loss

and ∆Pt

loss at initial operation. If ∆Pavg

loss and ∆Pt

loss riseagain after long-term operation of the power module, this canbe used to identify fatigue-related degradation of the packageor the power device that are reflected by the electrical orthermal behavior of the power module, e.g. bond wire lift-offs or solder delamination.

IV. ADAPTIVE SWITCHING LOSS MODEL CALIBRATION

This paper adds a loss injection unit and a model referenceadaptive system (MRAS) [87] to the observer for automaticswitching loss model calibration. This is necessary, due tovariation of the switching loss characteristics for differentdevices, gate drivers and modules parasitics. Conduction lossprediction is less critical, as it depends exclusively on thedevice such that off-line calibration or using data sheet valuesare effective options.

Typically, the gate resistance Rg and the dc-link voltage Vdcare not changed during the operation of the power module. Asa consequence, the averaged switching losses of the IGBT andthe diode are approximated, based on (10).

P IGBTsw ≈ K0

π· I · fpwm and PDiode

sw ≈ 0 (10)

The diode switching losses are approximated to zero, asthey are small compared to its conduction losses. Also, it istolerable assumption to neglect temperature dependencies aswell as current independent offsets in the IGBT switchinglosses.

PlossPloss

Ploss Tj

ReZ (w )th 0

ImZ (w )th 0

P sin(w t)0 0

P0 Ploss

Excitation and

Filtering

ThermalImpedance

Analysis

State of Health

Detection

Die Attach Solder

DCB Solder

Convection Process

w0

ex

ex ex

Rth,1

Cth,1

Rth,2

Cth,2

Rth,3

Cth,3

TaTj

Tj

T

Fig. 7. In-situ thermal impedance spectroscopy (ITIS)

The resulting expression (10) has a single tunable parameterK0. For identification of K0, switch A of the observer struc-ture in Fig. 4 is activated. This adds a rectangular excitationsignal ∆fsw(t) at a frequency of f1ex with an amplitude of∆fpwm to the operation switching frequency f0sw:

fsw = f0sw + ∆fsw(t) (11)

Consequently, any error of the estimated loss prediction∆P

avg

loss at this particular frequency f1ex can only occur dueto and incorrect K0. Thus, by correlation of the average lossprediction error ∆P

avg

loss with the excitation signal ∆fsw(t)and subsequent integration, a correction signal is generated.It automatically corrects K0, as it intrinsically drives theestimate of K0 to zero.

This self-calibration of the switching loss model is apowerful tool to obtain more accurate loss models duringinitial operation of the power converter avoiding sophisticatedoff-line calibrations. It makes the thermal monitoring moreaccurate at high bandwidth were the magnitude accuracy ofthe temperature and loss estimates depends on the loss model.

V. MONITORING LIFE-TIME VARYING THERMALPARAMETERS FOR DEGRADATION DIAGNOSIS

A. In-situ Thermal Impedance Spectroscopy (ITIS)

This section introduces in-situ thermal impedance spec-troscopy (ITIS) that extracts the thermal impedance of thepower module Zth(jω) over multiple frequency decades dur-ing normal converter operation. For ITIS, the observer isequipped with a resonant feedback path and operated togetherwith a device loss manipulation unit and a system identifica-tion algorithm.

The principle of the ITIS is illustrated in Fig. 7. In additionto the operation loss Ploss a small sinusoidal loss excitationP exex is injected in the devices that excites a temperature T ex

j .

P exloss = P0 · sin(ω0t) T ex

j = T0 · sin(ω0t+ ϕ) (12)

The load-independent loss excitation can be either generatedby modulation schemes that manipulate the zero sequenceduty cycle to control devices losses. Alternatively, an adaptivegate driver can be used that controls the switching processto manipulate losses. This work used an adaptive gate driverdeveloped in [63]. It uses multiple paralleled gate resistancesthat can be activated and deactivated every PWM period.A model-based loss manipulation algorithm ensures that thecommanded losses are realized by activating a suitable gateresistance combination.

T (k)j

P (k)0

P0

P0

2

2

ReZ (jw )(k)th 0

ImZ (jw )(k)th 0

2

2

P (k)0

Ploss

Ploss

eex

eexQ

w (k)0

sin(w t)0

cos(w t)0

ex

S

S

e (n)ex

e (n)ex

n = k-N

n = k-N

k

k

N

N

1

1

ex

ex Q

(k)

(k)

Fig. 8. Thermal impedance estimation via orthogonal correlation

The junction temperature response Tj that occurs due tothe load profile Ploss and the small signal loss excitationP exloss must be measured. In this work, the measurement is

realized with IR sensor that are placed above the devices of adecapsulated power module. However, in future applicationsTSEP sensing or integrated temperature sensors need to beappled for junction temperature measurement.

In a next step, the small-signal thermal response T exj must

be separated from the junction temperature caused by theoperational loss profile Ploss. This filtering is realized by anextension of the thermal observer structure in Fig. 4 thatis activated by the switch B. The activated third instanceof the thermal model (blue) is fed with the estimated lossexcitation P

ex

loss as a feedforward command, such that itideally estimates the resulting temperature excitation T

ex

j . Aresonant feedback loop, which is frequently used for currentcontrol of converters [67], [88], is tuned to the excitationfrequency ω0. It corrects any deviations of the temperatureestimation at the excitation frequency ω0 with zero amplitudeand phase error, as illustrated in the estimation accuracy plotin Fig. 5, even if the electrothermal model deviates from thephysical system. This observer-based filtering approach hasbeen developed by the authors in a strong analogy to observer-based filters for position self-sensing of electrical machinesthat were published in [89].

Finally, the isolated small-signal temperature T exj and its

excitation P exloss are processed via orthogonal correlation to

obtain the real and imaginary component of the thermalimpedance Zth(jω). The orthogonal correlation method is astandard tool for system identification that is illustrate in Fig.8 and discussed in [90]. By repeating this process for severaldiscrete values of ω0, the frequency response function (FRF)of the thermal impedance Zth(jω) is estimated over multiplefrequency decades. Different from state-of-the-art approachesthat require to interrupt normal converter operation for thermalimpedance extraction [91], this approach extracts the thermalimpedance without interrupting of converter operation. Theresulting information is very important for the state-of-healthdiagnosis of power modules, as it reveals localized degra-dation of the thermal interface of the power module, e.g.delamination or reduced convection [56], [92].

B. Utilizing Zth(jω)) for Degradation Diagnosis

The thermal impedance frequency response functionZth(jω)), which is extracted by ITIS, contains informationabout structural degradation of the power module, i.e. delam-inations of the device and substrate as well as deteriorationof the convection process. This intrinsically embedded infor-mation is illustrated in Fig. 9 showing the thermal-impedanceFRF of a IGBT in a healthy and differently degraded power

|Z(jw

)| in

K/W

th,j

∠(Z

(jw

)) in

°th

,j

Frequency in Hz

Thermal impedance of IGBTA

10-3

10-2

10-1

10-2 100 102 104

-45

0

-90

Normal

Convection Reduction

Die Delamination

DCB Solder Delamination

Fig. 9. Thermal impedance frequency response function plot representingthe thermal characteristics of the HP2 at different stages of degradation

modules. The thermal-impedance FRF is obtained followingthe modeling approach introduced in Section II. For a thermalmodel that includes delaminations, the thermal conductivity ofthe respective solder layer is reduced by 50 %. Similarly, forthe model with a deteriorated convection, the convection heattransfer coefficient is also reduced by 50 %. Both, magnitudeinformation |(Zth(jω))| and phase information ∠(Zth(jω))of the FRF plot reveal individual signatures of differentdegradation mechanisms. A detailed analysis and discussionof these degradation signatures has been conducted in [60]and [62].

Using the degradation signatures of thermal impedanceFRF for degradation diagnosis provides a key advantage.ITIS extracts the phase information arg(Zth(jω)) with zeroerror. This is because the precision of the phase extrac-tion only depends on accurate timing and is not influencedby moderate scaling errors of the measurement. Fig. 9, aswell as [22] show that phase information effectively allowsidentifying and separating different degradation mechanisms.Consequently, future monitoring system can strongly benefitfrom extracting phase information arg(Zth(jω)) with ITIScontinuously during normal converter operation. This enablesaccurate degradation diagnosis that shows minimal sensitivityto model errors.

VI. SIMULATION RESULTS

A. Adpative Switching Loss Model Calibration

First, instantaneous and averaged estimation of the devicelosses and the junction temperatures during adaptive calibra-tion of the loss model are investigated based on the simulationresults in Fig. 10. Initially, the switching loss parameter K0 isset to zero. The loss model calibration is activated at t = 22 s.This adds a rectangular excitation signal with a frequency of8 Hz and an amplitude of ∆fsw = 200 Hz to the nominalPWM frequency of fsw = 7 kHz. Without adaption, theswitching loss model is not accurate and loss deviations occur

T in

KJ

0 10 20 30 40 50 60 70 80-500 -50

0 0

500 50

1000 100

Lo

sse

s in

W

Lo

ss e

rro

r in

W

Losses

Loss Error

0 10 20 30 40 50 60 70 800

50

100

Instantaneous Temperature

Averaged Temperature

0 10 20 30 40 50 60 70 80

Time in s

-0.1

0

0.1

0.2

Ksw

in m

J/A

Estimated Switching loss gain of IGBTA

Fig. 10. Simulation of observer based IGBT junction temperature and deviceloss estimation with adaptive loss model calibration

that are estimated as a loss error by the observer. The averageloss error ∆Ploss corrects the estimation process within theobserver bandwidth and ensures the accurate estimation ofjunction temperature even before the loss model is calibrated.At its activation at t = 22 s, the adaption process startsusing the average loss estimation error ∆P avg

loss to adaptivelydetermine the correct switching loss gain K0. After a timehorizontal of 15 s the calibration process has converged andthe loss estimation error has decayed to zero.

B. In-situ Thermal Impedance Spectroscopy (ITIS)

In a second simulation, illustrated in Fig. 11, ITIS isevaluated during a mission load of an electric vehicle. Duringtransient converter operation, a periodic loss excitation withan amplitude of 30 W is generated by manipulation of thegate resistance [63] of IGBTA. The excitation frequencyis step-wise increased from 0.1 − 10 Hz to excite junctiontemperature response in this frequency range of the IGBT.The resulting instantaneous and averaged junction temperatureestimates are depicted in Fig. 11. They reflect dominantlythe mission load but also exhibit traces of the periodic lossexcitation. At low excitation frequencies one can identifythe temperature response to the excitation signal. At higherexcitations frequencies, the excitation losses are damped toostrongly by the thermal capacitance of the power module,such that the temperature response cannot be distinguishedby eye from the thermal profile that is caused by the missionprofile. However, the observer with its resonant feedbackpath is able to extract the thermal response T ex

j exclusivelyat the excitation frequency with great precision and zerophase lag. Based on this extracted thermal response T ex

j ,which is separated from the thermal response of the powermodule to the mission load, the orthogonal correlation unit canaccurately estimate magnitude and phase of IGBTA’s thermalimpedance at one excitation frequency after another.

0 20 40 60 80 100 120 140 160 180 200

0.1 Hz 0.2 Hz 0.5 Hz 1 Hz 2 Hz 5 Hz 10 Hz

-10

0

10

20F

reqeuncy

in H

z

-100

0

100

0 20 40 60 80 100 120 140 160 180 200-50

0

50

ex

ex

P in W

, T

in K

lo

ss

J

Estimated Losses

Estimated Temperature

0 20 40 60 80 100 120 140 160 180 20050

100

150

T in K

J

Time in s

Instantaneous Temperature

Averaged Temperature

0 20 40 60 80 100 120 140 160 180 2000

0.05

0.1

|Zth

| in

K/W

arg

(Zth

) in

°Magnitude

Phase

Fig. 11. Simulation of observer based in-situ thermal impedance spectroscopy(TIS) with 0.1 − 10 Hz excitation during transient converter operation

aidut

udut

Ri

Cdc Cdut

Lf

bidut

cidut

uidc

didc

Load Emualtor Device under Test (DUT)

udc

*Udc

3 Inductors (0.5mH/0.3mH) Driver Board

XCS 2000 Control Platform Hybridpack2 Power Module

Load emulator converter LT25F IR Sensors

Fig. 12. Experimental setup showing the decapsulated HP2 power modulewith adaptive gate drivers developed in [63] interfaced to a 3 phase loademulator [30]

VII. EXPERIMENTAL EVALUATION

The feasibility of the thermal monitoring concept is demon-strated for a HP2 power module on a three-phase load-emulator test bench, which is depicted in Fig. 12 and in-depth introduced in [30]. The power module was decapsu-lated such that the surface temperature at the center of thepower device could be measured with LT25F IR sensorsfrom Optris. This device surface temperature is used as avirtual junction temperature for the experimental evaluation.The power converter is operated with a gate driver developedin [63], that can dynamically adjust the gate resistances Rg

between 1.8 Ω and 18 Ω The monitoring, system excitationand identification algorithm are implemented in C++ on anXCS2005 rapid control prototyping platform from AixControland are operated with sampling frequency of fs = 1 kHz.

A. Thermal Monitoring

First, the basic functionality of the observer, i.e. monitoringdevice losses and temperatures within the power module, isevaluated over a realistic mission profile that is applied bythe load emulator and depicted in Fig. 13. The observer

Time in s

Tem

pe

ratu

re in

°C

Tem

pe

ratu

re in

°C

Ave

rag

ed

Lo

ss E

rro

r in

WL

osse

s in

WI

in

A, V

in V

, f in

°p

p

f in

Hz

e

0 10 20 30 40 50 60

0 10 20 30 40 50 60

0

200

400

0 10 20 30 40 50 60

20

25

30

0 1 2 3 4 5 6 7 8 9 10

20

25

30

0 10 20 30 40 50 60-20

0

20

40

IGBTA

Instantaneous

DiodeA

Averaged

IGBTA

DiodeA

IGBTA

DiodeA

IGBTA

DiodeA

Lateral Layer

Upper Sd

Lower Sd

Base Junction

Junction Temperature

Measurement

Averaged

Instantaneous

-100 -10

0 0

100 10

200 20

300 30Phase Current

Phase Voltage

Load angleExcitation frequency

Fig. 13. Experimental instanteneous and averaged loss and temperatureestimtion over a realistic load profile applied by a load emulator

Lo

sse

s in

W

Time in s Time in sK

/ K

sw

sw0 0

0 0

50 0.5

1001

20 2040 4060 60

Loss Estimation Error

Averaged Losses

Estimated gain

Low pass filtered gain

Fig. 14. Experimental response of the automatic loss model calibration

provides estimates of the instantaneous and averaged junc-tion temperature with excellent accuracy while rejecting thehigh bandwidth measurement noise. Furthermore, it providesspatial temperature estimates of the upper and lower solderlayer temperature and the temperature of the base plate. Inaddition, the observer provides closed-loop estimates of thedevice losses. The averaged loss-error reflects the part of thelosses that were not accurately predicted by the device lossesmodel, which exhibits deviations of up to 5−15 % dependentof the operation point.

B. Adaptive Loss Model Calibration

The device loss model can be calibrated by adaptive esti-mation of the switching loss parameter K0 in Fig. 14. Duringthe adaptive parameter estimation, the power module wasoperated with a current of 300 A, a constatn dc-link voltageof 100 V, an AC voltage amplitude of 10 V, a load angle of10 °, a frequency of fbase = 20 Hz and an average switchingfrequency of f0sw = 7 kHz. The excitation signal had a fre-quency of f1ex = 4 Hz and an amplitude of ∆fpwm = 500 Hz.The feedback gain of the adaptive parameter estimation wasset to km = 0.3 ns/A2. Initially, the loss parameter K0 wasset to 50 % of its true value. It can be seen in Fig. 14 that

8.5 9 9.5 10 10.5 11 11.5 12 12.5

-10

0

10

20

Loss

in W

, R

esi

stance

in

Excitation

Quadrature Excitation

Rgate

0 5 10 15 20 25 30 35 40 45 50

0

20

40

Tem

pera

ture

in °

C

Instantaneous temperature estimate

Averaged temperature estimate

Estimated Tj response

0 5 10 15 20 25 30 35 40 45 500

0.02

0.04

0.06

|Zth

| in

K/W

0 5 10 15 20 25 30 35 40 45 50

Time in s

-100

-50

0

50

arg

(Zth

) in

K/W

48

44

40

19.8 20 20.2

j

Fig. 15. Thermal impedance estimation of a IGBT during normal converteroperation at a current of 300 A, a dc-link voltage of 100 V, an AC voltageamplitude of 20 V, a load angle of 10 °, a frequency of fbase = 20 Hz anda PWM frequency of f0sw = 7000 Hz.

over a time horizon of 70 s the switching loss parameter K0

is adaptively corrected to its original value. The averaged lossestimation error simultaneously decays to zero.

C. Thermal Impedance Spectroscopy

Finally, ITIS is experimentally evaluated during converteroperation with a current of 350 A, a dc-link voltage of 100 V,a phase voltage of 20 V, a load angle of 15 °, a fundamentalfrequency of 50 Hz and a switching frequency of the 5 kHz.The thermal impedance extraction process at an electrothermalexcitation frequency of 1 Hz is illustrated in Fig. 15. Att = 8 s, the IGBT losses are excited with an amplitude of15 W via gate resistance manipulation between Rg = 3 Ω andRg = 12 Ω. This causes a very small temperature responseT exj whose ∆T = 1.5 K is a factor of two smaller than the

device temperature variation caused by the current alternation.The observer is able to separate this response T ex

j from theremaining junction temperature signal, such that it can beprocessed by the orthogonal correlation unit to compute thethermal impedance Zth(jω0) in magnitude and phase. Alluncorrelated noise gets rejected by the averaging processwithin the orthogonal correlation unit resulting in a smoothestimate of the thermal impedance.

This process has been repeated for excitation frequenciesfrom 50 mHz to 5 Hz to determine the FRF of the thermalimpedance Zth(jω) that is plotted in Fig. 16 in comparison tothe thermal model developed in II. The extracted FRF of theZth exhibits a maximum deviation of 13 % in magnitude and2 ° in phase from the model. This shows that the proposed ITISconcept is able to extract the Zth FRF over a wide frequency

10-2

10-1

|Zth

(jw

)| in K

/W

On-line extracted Zth

Model based Zth

10-1 100 101

Frequency in Hz

-80

-60

-40

-20

0

arg

(Zth

(jw

)) in K

/W

Fig. 16. Frequency response function (FRF) of the thermal impedanceZth(jω) extracted from a accurate model that was calibrated based onexperimental Zth(t) measurements [41] and FRF extracted via ITIS byrepetition of the impedance estimation illustrated in Fig. 15 at a wide rangeof excitation frequencies

range with excellent precision without interrupting converteroperation.

The frequency range in which the proposed ITIS conceptworks is limited by the bandwidth, signal-to-noise ratio (SNR)and the resolution of the utilized temperature sensing tech-nology. In the presented experimental work, the frequencyrange was limited to 10 Hz, as the IR sensors provided noisydata at higher bandwidth. Consequently, there is a strong de-mand integrated high-bandwidth junction temperature sensingtechnologies, e.g. based on temperature sensitive electrical[34], [75], [77], [79] or optical parameters [80], [93].Thetechnologies to be used should be selected based on theirbandwidth, SNR, temperature resolution and insensitivity todisturbances.

A second limit to the frequency range of the proposed ITISconcept lies in the decaying characteristics of the thermalimpedance at higher excitation frequencies. This requires toincrease the loss excitation amplitude for extracting thermalimpedances at higher bandwidth, such that the resulting tem-perature response has an amplitude that is sufficiently large tobe measured. As the loss manipulation ability of active gatedrivers, which were used in this work, but also of other lossmanipulation techniques is limited, this results in a maximalbandwidth above which the thermal impedance cannot bedetected. However, this is not only a limit of the proposedtechnology, but a general limit for all thermal impedanceextraction technologies.

VIII. CONCLUSIONS

The contributions and key associated conclusions of thispaper are summarized as follows:

• Real-time calibration and extraction of life-time varyingparameters of the power module without interruptingnormal converter operation

– Observer correction variables allow detecting elec-trothermal model errors as well as changes of theelectrothermal behavior, e.g. due to degradation

– Self-correction of errors in the loss model with amodel reference adaptive system

– In-situ thermal impedance spectroscopy extractsthermal impedance with minimal magnitude andzero phase error over multiple frequency decades

• Thermal monitoring, i. e. 3-D temperature and lossestimation, with an observer that uses a self-calibratingdevice loss model yields improved accuracy even at highbandwidth and provides:

– Thermal overload protection of the power module atrapidly changing loading

– Closed-loop loss estimation of individual devices– Detection of hard- and soft switching operation

• Identification of frequency-dependent thermal impedanceZth(jω) as key indicator of power module state-of-health

– Phase information is highly sensitive to structuralchanges of the power module and can be extractedwith superior accuracy compared to magnitude

– Strong opportunity for future diagnosis systems thatevaluate phase information instead of magnitude

• Experimental evaluation of the thermal monitoring sys-tem on a state-of-the-art IGBT power module duringrealistic converter operation accomplishes

– Closed loop monitoring of instantaneous and aver-aged spatial temperatures and device losses

– Successful loss model parameter identification– Accurate extraction of thermal impedance transfer

function Zth(jω) over multiple frequency decades

ACKNOWLEDGMENT

The authors would like to thank the German AcademicExchange Service (DAAD), the Institute of Power Electronicsand Electrical Drives (ISEA) at RWTH Aachen University andthe Wisconsin Electric Machines and Power Electronics Con-sortium (WEMPEC) at the University of Wisconsin-Madisonfor supporting this research.

REFERENCES

[1] C. Neeb, L. Boettcher, M. Conrad, and R. De Doncker, “Innovativeand reliable power modules: A future trend and evolution of technolo-gies,” Industrial Electronics Magazine, IEEE, vol. 8, no. 3, pp. 6–16,Sep. 2014.

[2] A. H. Wienhausen, A. Sewergin, and R. W. De Doncker, “Highlyintegrated two-phase SiC boost converter with 3D-printed fluidcoolers and 3D-printed inductor bobbins,” in International Exhibitionand Conference for Power Electronics, Intelligent Motion, RenewableEnergy and Energy Management (PCIM), 2018.

[3] A. Stippich, C. H. van der Broeck, A. Sewergin, A. H. Wienhausen,M. Neubert, P. Schülting, S. Taraborrelli, H. V. Hoek, and R. W.De Doncker, “Key components of modular propulsion systems fornext generation electric vehicles,” CPSS Transactions on PowerElectronics and Applications, vol. 2, no. 4, pp. 249–258, Dec. 2017.

[4] K. Wang, “Review of state-of-the-art integration technologies inpower electronic systems,” CPSS Transactions on Power Electronicsand Applications, vol. 2, no. 4, pp. 292–305, Dec. 2017.

[5] S. Yang, D. Xiang, A. Bryant, P. Mawby, L. Ran, and P. Tavner,“Condition monitoring for device reliability in power electronicconverters: A review,” IEEE Transactions on Power Electronics,vol. 25, no. 11, pp. 2734–2752, Nov. 2010.

[6] F. B. Henry Shu-hung Chung Huai Wang and M. Pecht, Reliabilityof Power Electronic Converter Systems. IET Digital Library, 2016.

[7] M. Andresen, K. Ma, G. Buticchi, J. Falck, F. Blaabjerg, andM. Liserre, “Junction temperature control for more reliable powerelectronics,” IEEE Transactions on Power Electronics, vol. 33, no. 1,pp. 765–776, Jan. 2018.

[8] D. A. Murdock, J. E. R. Torres, J. J. Connors, and R. D. Lorenz,“Active thermal control of power electronic modules,” IEEE Trans-actions on Industry Applications, vol. 42, no. 2, pp. 552–558, Mar.2006.

[9] M. A. Eleffendi and C. M. Johnson, “In-service diagnostics for wire-bond lift-off and solder fatigue of power semiconductor packages,”IEEE Transactions on Power Electronics, vol. 32, no. 9, pp. 7187–7198, Sep. 2017.

[10] T. Polom, C. H. van der Broeck, R. W. De Doncker, and R. D. Lorenz,“Real-time, in situ degradation monitoring in power semiconductorconverters,” in 2018 IEEE Applied Power Electronics Conference andExposition (APEC), 2019.

[11] H. Soliman, H. Wang, and F. Blaabjerg, “A review of the condi-tion monitoring of capacitors in power electronic converters,” IEEETransactions on Industry Applications, vol. 52, no. 6, pp. 4976–4989,Nov. 2016.

[12] P. Sundararajan, M. H. M. Sathik, F. Sasongko, C. S. Tan, M. Tariq,and R. Simanjorang, “Online condition monitoring system for DC-link capacitor in industrial power converters,” IEEE Transactions onIndustry Applications, vol. 54, no. 5, pp. 4775–4785, Sep. 2018.

[13] T. A. Polom, B. Wang, and R. D. Lorenz, “Control of junctiontemperature and its rate of change at thermal boundaries via preciseloss manipulation,” IEEE Transactions on Industry Applications,vol. 53, no. 5, pp. 4796–4806, Sep. 2017.

[14] C. H. van der Broeck, L. A. Ruppert, R. D. Lorenz, and R. W. A. DeDoncker, “Methodology for active thermal cycle reduction of powerelectronic modules,” IEEE Transactions on Power Electronics, 2018.

[15] M. A. Eleffendi and C. M. Johnson, “Application of kalman filterto estimate junction temperature in igbt power modules,” IEEETransactions on Power Electronics, vol. 31, no. 2, pp. 1576–1587,Feb. 2016.

[16] M. Musallam and C. M. Johnson, “Real-time compact thermal modelsfor health management of power electronics,” IEEE Transactions onPower Electronics, vol. 25, no. 6, pp. 1416–1425, Jun. 2010.

[17] M. Andresen, M. Schloh, G. Buticchi, and M. Liserre, “Computa-tional light junction temperature estimator for active thermal control,”in 2016 IEEE Energy Conversion Congress and Exposition (ECCE),2016.

[18] C. H. van der Broeck, R. D. Lorenz, and R. W. A. De Doncker,“Monitoring 3-d temperature distributions and device losses in powerelectronic modules,” IEEE Transactions on Power Electronics, 2018.

[19] D. Kaczorowski, B. Michalak, and A. Mertens, “A novel thermalmanagement algorithm for improved lifetime and overload capabil-ities of traction converters,” in EPE’15 ECCE-Europe, Sep. 2015,pp. 1–10.

[20] C. H. van der Broeck and R. W. De Doncker, “Thermal monitoringof power electronic modules with minimal sensing effort,” in 2019IEEE Energy Conversion Congress and Exposition (ECCE), 2019.

[21] J. Woelfle, O. Lehmann, and J. Roth-Stielow, “A novel controlmethod to improve the reliability of traction inverters for permanentmagnet synchronous machines,” in 2015 IEEE PEDS, Jun. 2015,pp. 379–384.

[22] T. A. Polom, C. H. van der Broeck, D. D. R. W., and L. R. D.,“Spatially-varying electrothermal impedance analysis for designingpower semiconductor converter systems,” in Intersociety Conferenceon Thermal and Thermomechanical Phenomena in Electronic Sys-tems (ITherm), 2019.

[23] J. Poon, P. Jain, C. Spanos, S. K. Panda, and S. R. Sanders, “Faultprognosis for power electronics systems using adaptive parameteridentification,” IEEE Transactions on Industry Applications, vol. 53,no. 3, pp. 2862–2870, May 2017.

[24] U. M. Choi and F. Blaabjerg, “Separation of wear-out failure modesof igbt modules in grid-connected inverter systems,” IEEE Transac-tions on Power Electronics, vol. 33, no. 7, pp. 6217–6223, Jul. 2018.

[25] M. Liserre, M. Andresen, L. Costa, and G. Buticchi, “Power routingin modular smart transformers: Active thermal control through unevenloading of cells,” IEEE Industrial Electronics Magazine, vol. 10,no. 3, pp. 43–53, Sep. 2016.

[26] Y. Avenas, L. Dupont, N. Baker, H. Zara, and F. Barruel, “Conditionmonitoring: A decade of proposed techniques,” IEEE IndustrialElectronics Magazine, vol. 9, no. 4, pp. 22–36, Dec. 2015.

[27] C. van der Broeck, S. Kalker, and R. De DOncker, “Overloadand degradation monitoring of power electronic converters,” in 11thExpert Forum Electric Vehicle Drives, E-MOTIVE 2019, 2019-09-04- 2019-09-05, Schweinfurt, Germany, 2019.

[28] X. Wang, A. Castellazzi, and P. Zanchetta, “Observer based dynamicadaptive cooling system for power modules,” Microelectronics Re-liability, vol. 58, pp. 113–118, 2016, Reliability Issues in PowerElectronics.

[29] C. H. van der Broeck, R. D. Lorenz, and R. W. De Doncker, “Methodsfor monitoring 3-d temperature distributions in power electronicmodules,” in 2018 IEEE Applied Power Electronics Conference andExposition (APEC), IEEE, Mar. 2018.

[30] C. H. van der Broeck, H. Zeng, R. D. Lorenz, and R. W. De Doncker,“A test bench for thermal characterization of igbt power modules overmission profiles,” in PCIM Europe 2018; International Exhibitionand Conference for Power Electronics, Intelligent Motion, RenewableEnergy and Energy Management, Jun. 2018.

[31] R. Lorenz and K. V. Patten, “High-resolution velocity estimationfor all-digital, AC servo drives,” IEEE Transactions on IndustryApplications, vol. 27, no. 4, pp. 701–705, 1991.

[32] P. L. Jansen, “The integration of state estimation, control and designfor induction machines,” PhD thesis, UW Madison, 1994.

[33] C. H. van der Broeck, R. W. De Doncker, S. A. Richter, andJ. V. Bloh, “Unified control of a buck converter for wide-load-rangeapplications,” IEEE Transactions on Industry Applications, vol. 51,no. 5, pp. 4061–4071, Sep. 2015.

[34] N. Baker, M. Liserre, L. Dupont, and Y. Avenas, “Improved reliabilityof power modules: A review of online junction temperature measure-ment methods,” IEEE Industrial Electronics Magazine, vol. 8, no. 3,pp. 17–27, Sep. 2014.

[35] J. Winkler, J. Homoth, and I. Kallfass, “Electroluminescence-basedjunction temperature measurement approach for sic power mosfets,”IEEE Transactions on Power Electronics, pp. 1–1, 2019.

[36] E. R. Motto and J. F. Donlon, “Igbt module with user accessibleon-chip current and temperature sensors,” in 2012 Twenty-SeventhAnnual IEEE Applied Power Electronics Conference and Exposition(APEC), Feb. 2012, pp. 176–181.

[37] T. K. Gachovska, B. Tian, J. L. Hudgins, W. Qiao, and J. F. Donlon,“A real-time thermal model for monitoring of power semiconductordevices,” IEEE Transactions on Industry Applications, vol. 51, no. 4,pp. 3361–3367, Jul. 2015.

[38] J. Reichl, J. Ortiz-Rodriguez, A. Hefner, and J.-S. Lai, “3-d thermalcomponent model for electrothermal analysis of multichip powermodules with experimental validation,” Power Electronics, IEEETransactions on, vol. 30, no. 6, pp. 3300–3308, Jun. 2015.

[39] A. Antoulas and S. D.C., “Approximation of large-scale dynamicalsystems: An overview,” International J. of Applied Mathematics andComputational Science, vol. 11, pp. 1093–1121, 2001.

[40] F. Qi, D. A. Ly, C. H. van der Broeck, D. Yan, and R. W. DeDoncker, “Model order reduction suitable for online linear parameter-varying thermal models of electric motors,” in 2016 IEEE 2nd AnnualSouthern Power Electronics Conference (SPEC), IEEE, Dec. 2016,pp. 1–6.

[41] C. H. van der Broeck, M. Conrad, and R. W. De Doncker, “Athermal modeling methodology for power semiconductor modules,”Microelectronics Reliability, vol. 55, no. 9-10, pp. 1938–1944, Aug.2015.

[42] C. H. van der Broeck, L. A. Ruppert, A. Hinz, M. Conrad, andR. W. De Doncker, “Spatial electro-thermal modeling and simula-tion of power electronic modules,” IEEE Transactions on IndustryApplications, vol. 54, no. 1, pp. 404–415, Jan. 2018.

[43] R. Goldbeck, C. H. van der Broeck, and R. W. De Doncker, “Electro-thermal simulation of bond wires in power modules for realisticmission profiles,” in 12th IEEE International Conference on PowerElectronics and Drive Systems (PEDS 2017), IEEE, Dec. 2017.

[44] X. Wang, A. Castellazzi, and P. Zanchetta, “Observer based tempera-ture control for reduced thermal cycling in power electronic cooling,”Applied Thermal Engineering, vol. 64, no. 1–2, pp. 10–18, 2014.

[45] M. H. M. Sathik, F. S. Sundararajan Prasanth, S. K. Padmanabhan,J. Pou, and R. Simanjorang, “A dynamic thermal controller for powersemiconductor devices,” in Applied Power Electronics Conferenceand Exposition (APEC), 2018.

[46] C. Li, D. Jiao, J. Jia, F. Guo, and J. Wang, “Thermoelectric cooling forpower electronics circuits: Modeling and active temperature control,”IEEE Transactions on Industry Applications, vol. 50, no. 6, pp. 3995–4005, Nov. 2014.

[47] A. S. Bahman, K. Ma, and F. Blaabjerg, “A lumped thermal modelincluding thermal coupling and thermal boundary conditions for high-power IGBT modules,” IEEE Transactions on Power Electronics,vol. 33, no. 3, pp. 2518–2530, Mar. 2018.

[48] Z. Hu, M. Du, K. Wei, and W. G. Hurley, “An adaptive thermalequivalent circuit model for estimating the junction temperature ofIGBT’s,” IEEE Journal of Emerging and Selected Topics in PowerElectronics, pp. 1–1, 2018.

[49] K. Ammous, H. Morel, and A. Ammous, “Analysis of powerswitching losses accounting probe modeling,” IEEE Transactions onInstrumentation and Measurement, vol. 59, no. 12, pp. 3218–3226,Dec. 2010.

[50] Z. Zhang, B. Guo, F. F. Wang, E. A. Jones, L. M. Tolbert, andB. J. Blalock, “Methodology for wide band-gap device dynamiccharacterization,” IEEE Transactions on Power Electronics, vol. 32,no. 12, pp. 9307–9318, Dec. 2017.

[51] J. Gottschlich, M. Kaymak, M. Christoph, and R. De Doncker, “Aflexible test bench for power semiconductor switching loss measure-ments,” in 11th International Conference on Power Electronics andDrive Systems, Jun. 2015, pp. 442–448.

[52] X. Li, L. Zhang, S. Guo, Y. Lei, A. Q. Huang, and B. Zhang,“Understanding switching losses in SiC MOSFET: Toward losslessswitching,” in 2015 IEEE 3rd Workshop on Wide Bandgap PowerDevices and Applications (WiPDA), IEEE, Nov. 2015.

[53] G. Engelmann, N. Fritz, C. Lüdecke, and R. W. De Doncker, “Impactof the different parasitic inductances on the switching behavior of SiCMOSFETs,” in IEEE International Power Electronics and MotionControl Conference (PEMC), IEEE, Aug. 2018, pp. 918–925.

[54] D. Xiang, L. Ran, P. Tavner, A. Bryant, S. Yang, and P. Mawby,“Monitoring solder fatigue in a power module using case-above-ambient temperature rise,” IEEE Transactions on Industry Applica-tions, vol. 47, no. 6, pp. 2578–2591, Nov. 2011.

[55] M. A. Eleffendi and C. M. Johnson, “Thermal path integrity mon-itoring for igbt power electronics modules,” in CIPS 2014; 8thInternational Conference on Integrated Power Electronics Systems,Feb. 2014, pp. 1–7.

[56] F. Grieger and A. Lindemann, “Thermal impedance spectroscopyfor non-destructive evaluation of power cycling,” in 2015 IEEE6th International Symposium on Power Electronics for DistributedGeneration Systems (PEDG), Jun. 2015, pp. 1–6.

[57] A. Hensler, D. Wingert, C. Herold, J. Lutz, and M. Thoben, “Thermalimpedance spectroscopy of power modules during power cycling,” in2011 IEEE 23rd International Symposium on Power SemiconductorDevices and ICs, IEEE, May 2011, pp. 264–267.

[58] A. M. Aliyu and A. Castellazzi, “Prognostic system for powermodules in converter systems using structure function,” IEEE Trans-actions on Power Electronics, vol. 33, no. 1, pp. 595–605, Jan. 2018.

[59] M. A. Eleffendi, L. Yang, P. Agyakwa, and C. M. Johnson, “Quan-tification of cracked area in thermal path of high-power multi-chipmodules using transient thermal impedance measurement,” Micro-electronics Reliability, vol. 59, pp. 73–83, Apr. 2016.

[60] C. H. van der Broeck, S. Kalker, T. A. Polom, W. De Doncker, andL. R. D., “In-situ thermal impedance spectroscopy of power electronicmodules for localized degradation identification,” in InternationalExhibition and Conference for Power Electronics, Intelligent Motion,Renewable Energy and Energy Management (PCIM Europe), 2019.

[61] W. Lai, M. Chen, L. Ran, S. Xu, N. Jiang, X. Wang, O. Alatise, and P.Mawby, “Experimental investigation on the effects of narrow junctiontemperature cycles on die-attach solder layer in an igbt module,”IEEE Transactions on Power Electronics, vol. 32, no. 2, pp. 1431–1441, Feb. 2017.

[62] T. A. Polom, C. H. van der Broeck, R. W. A. De Doncker, andR. D. Lorenz, “Real-time, in situ degradation monitoring in powersemiconductor converters,” in IEEE Applied Power Electronics Con-ference and Exposition (APEC), 2019.

[63] C. H. van der Broeck, L. A. Ruppert, R. D. Lorenz, and R. W. DeDoncker, “Active thermal cycle reduction of power modules via gateresistance manipulation,” in 2018 IEEE Applied Power ElectronicsConference and Exposition (APEC), IEEE, Mar. 2018.

[64] C. H. van der Broeck, T. Polom, R. D. Lorenz, and R. W. De Doncker,“Thermal monitoring of power electronic modules via device self-sensing,” in 2018 IEEE Energy Conversion Congress and Exposition(ECCE), Sep. 2018.

[65] C. H. van der Broeck, S. Kalker, T. A. Polom, W. De Doncker, andL. R. D., “In-situ thermal impedance spectroscopy of power electronicmodules for localized degradation identification,” in International

Exhibition and Conference for Power Electronics, Intelligent Motion,Renewable Energy and Energy Management (PCIM Europe), 2019.

[66] R. Isermann, Digital Control Systems Volume 1. Springer Press, 1989.[67] C. H. van der Broeck, R. W. De Doncker, S. A. Richter, and J.

von Bloh, “Discrete time modeling, implementation and design ofcurrent controllers,” in 2014 IEEE Energy Conversion Congress andExposition (ECCE), IEEE, Sep. 2014, pp. 540–547.

[68] FS800R07A2E3B13, Hybridpack2 data sheet, Infineon TechnologiesAG, 2011.

[69] L. Dupont, Y. Avenas, and P. O. Jeannin, “Comparison of junctiontemperature evaluations in a power igbt module using an ir cameraand three thermosensitive electrical parameters,” IEEE Transactionson Industry Applications, vol. 49, no. 4, pp. 1599–1608, Jul. 2013.

[70] G. Zeng, H. Cao, W. Chen, and J. Lutz, “Difference in devicetemperature determination using pn-junction forward voltage and gatethreshold voltage,” IEEE Transactions on Power Electronics, pp. 1–1,2018.

[71] M. Schubert and R. W. De Doncker, “Semiconductor temperature andcondition monitoring using gate driver integrated inverter output volt-age measurement,” in International Symposium on Power Electronics,Electrical Drives, Automation and Motion (SPEEDAM), IEEE, Jun.2018.

[72] L. Dupont and Y. Avenas, “Preliminary evaluation of thermo-sensitiveelectrical parameters based on the forward voltage for online chiptemperature measurements of igbt devices,” IEEE Transactions onIndustry Applications, vol. 51, no. 6, pp. 4688–4698, Nov. 2015.

[73] F. Stella, G. Pellegrino, E. Armando, and D. Dapra, “On-line junctiontemperature estimation of sic power mosfets through on-state voltagemapping,” IEEE Transactions on Industry Applications, vol. PP,no. 99, pp. 1–1, 2018.

[74] N. Baker, S. Munk-Nielsen, F. Iannuzzo, and M. Liserre, “Igbtjunction temperature measurement via peak gate current,” IEEETransactions on Power Electronics, vol. 31, no. 5, pp. 3784–3793,May 2016.

[75] H. Niu and R. D. Lorenz, “Evaluating different implementations ofonline junction temperature sensing for switching power semicon-ductors,” IEEE Transactions on Industry Applications, vol. 53, no. 1,pp. 391–401, Jan. 2017.

[76] H. Kuhn and A. Mertens, “On-line junction temperature measurementof igbts based on temperature sensitive electrical parameters,” in 200913th European Conference on Power Electronics and Applications,Sep. 2009, pp. 1–10.

[77] S. Kalker, C. H. van der Broeck, and R. W. De Doncker, “Onlinejunction-temperature sensing of sic mosfets with minimal calibrationeffort,” in PCIM Europe 2020; International Exhibition and Confer-ence for Power Electronics, Intelligent Motion, Renewable Energyand Energy Management, May 2020.

[78] M. Denk and M. M. Bakran, “An igbt driver concept with integratedreal-time junction temperature measurement,” in PCIM Europe 2014;International Exhibition and Conference for Power Electronics, In-telligent Motion, Renewable Energy and Energy Management, May2014, pp. 1–8.

[79] C. H. van der Broeck, A. Gospodinov, and R. W. De Doncker, “IGBTjunction temperature estimation via gate voltage plateau sensing,”IEEE Transactions on Industry Applications, pp. 1–1, Sep. 2018.

[80] J. Winkler, J. Homoth, and I. Kallfass, “Electroluminescence-basedjunction temperature measurement approach for SiC power MOS-FETs,” IEEE Transactions on Power Electronics, vol. 35, no. 3,pp. 2990–2998, Mar. 2020.

[81] L. Ceccarelli, H. Luo, and F. Iannuzzo, “Investigating SiC MOSFETbody diode’s light emission as temperature-sensitive electrical pa-rameter,” Microelectronics Reliability, vol. 88-90, pp. 627–630, Sep.2018.

[82] C. Li, Z. Lu, H. Wu, W. Li, and X. He, “Junction temperaturemeasurement based on electroluminescence effect in body diodeof sic power mosfet,” in 2019 IEEE Applied Power ElectronicsConference and Exposition (APEC), 2019.

[83] S. Kalker, C. H. van der Broeck, and De Doncker, “Utilizing elec-troluminescence of sic mosfets for unified junction-temperature andcurrent sensing,” in 2020 IEEE Applied Power Electronics Conferenceand Exposition (APEC), 2020.

[84] C. Kempiak, A. Lindemann, E. Thal, and S. Idaka, “Investigation ofthe usage of a chip integrated sensor to determine junction temper-ature during power cycling tests,” in CIPS 2018; 10th InternationalConference on Integrated Power Electronics Systems, Mar. 2018,pp. 1–6.

[85] M. K. Kim and S. W. Yoon, “Miniature piezoelectric sensor for in-situtemperature monitoring of silicon and silicon carbide power modulesoperating at high temperature,” IEEE Transactions on Industry Ap-plications, vol. 54, no. 2, pp. 1614–1621, Mar. 2018.

[86] R. D. Lorenz and S. P.B., “Combining drives of different bandwidthsto meet process objectives,” WEMPEC, UW-Madison, WEMPECResearch Reports, Feb. 1989.

[87] K. Astrom and B. Wittenmark, Adaptive Control. Addison -Wesley,1989.

[88] C. H. van der Broeck, S. A. Richter, J. von Bloh, and R. W. D. Don-cker, “Methodology for analysis and design of DiscreteTime currentcontrollers for three-phase PWM converters,” CPSS Transactions onPower Electronics and Applications, vol. 3, no. 3, pp. 254–264, Sep.2018.

[89] M. S. Petit, B. Sarlioglu, R. D. Lorenz, and C. H. van der Broeck,“Carrier separation techniques for improved disturbance rejection ofinjection-based self-sensing control,” in 2019 IEEE 10th InternationalSymposium on Sensorless Control for Electrical Drives (SLED),IEEE, Sep. 2019.

[90] R. Isermann and M. Münchhof, Identification of Dynamic Systems- An Introduction with Applications, 1st ed. Springer-Verlag BerlinHeidelberg, 2011.

[91] A. M. Aliyu, S. Chowdhury, and A. Castellazzi, “In-situ healthmonitoring of power converter modules for preventive maintenanceand improved availability,” in 2015 17th European Conference onPower Electronics and Applications (EPE’15 ECCE-Europe), IEEE,Sep. 2015.

[92] M. Denk, M. M. Bakran, and S. Schafferhans, “Case sensitivecondition monitoring of an igbt inverter in a hybrid car,” in CIPS2016; 9th International Conference on Integrated Power ElectronicsSystems, Mar. 2016, pp. 1–6.

[93] S. Kalker, C. H. van der Broeck, and De Doncker, “Utilizingelectroluminescence of sic mosfets for unified junction-temperatureand current sensing,” in IEEE Applied Power Electronics Conferenceand Exposition (APEC), 2020.