Fault detection by means of Hilbert Huang
Transform of the stator current in a PMSM with
demagnetization
J. Rosero, A. Garcia, L. Romeral, J.A. Ortega
Motion Control and Industrial Applications Group, Universitat Politècnica de Catalunya.
C/ Colom 1, TR 2, 08222 Terrassa, Spain, e-mail: romeral@eel.upc.edu
Abstract – This paper presents a study of the permanent
magnet synchronous motor (PMSM) running under
demagnetization. The simulation has been carried out by means
of two dimensional (2-D) finite element analysis (FEA), and
simulations results were compared with experimental results.
The demagnetization fault is analyzed by means of decomposition
of stator currents obtained at different speeds. The Hilbert
Huang transform (HHT) is used as processing tool. This
transformation represents time-dependent series in a two-
dimensional (2-D) time-frequency domain by extracting
instantaneous frequency components within the signal through
an Empirical Mode Decomposition (EMD) process.
Index Terms – Drive, PMSM, fault, current, demagnetization,
empirical mode decomposition, Hilbert Huang transform,
simulation, FEA.
I. INTRODUCTION.
Permanent magnet synchronous machines (PMSM) are
attractive for a variety of applications because of their high
power density, wide constant-power speed range, and
excellent efficiency [1, 2]. However, faults in either the
machine or inverter create special challenges in any type of
PM synchronous machine drive because of the presence of
spinning rotor magnets that cannot be turned off at will.
The finite element analysis (FEA), which allows the
coupling between the non-linear magnetic and electric circuits,
is took into account for the motor analysis to predict the
performance characteristics of a system, especially when the
system is under fault conditions [3].
FFT of the stator current has been applied to detect
demagnetization faults by analyzing specific harmonics.
However, it can not be applied to non-stationary signals.
Moreover, FFT analysis can not differentiate harmonics due to
demagnetization from others due to eccentricity [1, 4].
Time-frequency analysis methods have been used for non-
stationary signal feature extraction, although successful
application of these techniques requires understanding of their
respective limitations. The selection of a suitable window size
is required when applying the Short-Time Fourier Transform
(STFT) to match with the specific frequency content of the
signal, which is generally not known a priori. A very
appealing feature of the continuous wavelet analysis (CWT) is
that it provides a uniform resolution for all the scales [5].
Limited by the size of the basic wavelet function, the
downside of the uniform resolution is a uniformly poor
resolution. Moreover, an important limitation of the wavelet
analysis is its non-adaptive nature. Once the basic wavelet is
selected, it is used to analyze the whole frequency range [6]. A
basic time-frequency representation is done by the Wigner-
Ville distribution (WVD), which is part of the Cohen class of
distribution [7]. The difficulty with this method is the severe
cross terms as indicated by the existence of negative power for
some frequency ranges. In addition, the WVD of discrete time
signals suffers from the aliasing problem.
The Hilbert–Huang transform (HHT) is based on the
instantaneous frequencies resulting from the Intrinsic Mode
Functions (IMF) of the signal being analyzed [8]; thus, it is
not constrained by the uncertainty limitations with respect to
the time and frequency resolutions to which other time-
frequency techniques are subject. In recent years, HHT has
been applied to identification of damage time instant and
location in civil and mechanical structures.
In this paper, a PMSM with demagnetization is simulated at
different speed values using a two-dimensional (2-D) finite
element analysis (FEA). A novel approach for nonlinear, non-
stationary data analysis by application of Hilbert-Huang
transform method is presented. The method allows a univocal
detection of the demagnetization failure. Simulations results,
which are corroborate by experiments too, demonstrate that
this method can effectively diagnoses the fault of
demagnetization.
II. SIMULATION OF PMSM WITH DEMAGNETIZATION
PERMANENT MAGNET.
Numerical simulations were developed over a model
obtained with the combination of a finite element software,
Flux2D [9] for the motor model, and Matlab-Simulink for
electronics and control. Both circuits, electromagnetic and
electric, have been coupled automatically by linking local
variations in flux with the circuit voltage as shows in Fig. 1.
The demagnetization can provoke irreversible losses that
cause the operating point to ``fall off`` at the lower end of a
recoil line, then appearing an irreversible flux loss [10] as
shows in the Fig. 2. The permanent magnets can be
demagnetized by fault currents such as short circuit currents
produced by inverter or stator faults, with an over-running
load, or where two motors are coupled to a single load.
Moreover, short circuit of the winding can be troublesome
because of drag torque and potential overheating of the motor
[4].
As a consequence of demagnetization, the distributed
magnetomotive force (MMF) is not sinusoidal. Thus, the
amplitude of MMF takes on constant amplitude less that
change in the pair of poles below the fault (Fig. 3).
Fig. 1. Schematic of the motor and electronics model for a PMSM
Fig. 2. Demagnetization curve of a permanent magnet (PM)
Fig. 3. Flux density distribution in air gap of a PMSM with 50 % magnetized.
Thus, the MMF under failure together with the constant
permeance induces currents of main multiple frequency. If
demagnetization exists, low-frequency components near the
fundamental appear [11], given by
f dmg = f s 1 ± k / p k = 1,2,3,... (1)
In case of constant speed at high and medium speeds, FFT
allows detecting demagnetization fault by only analyzing the
amplitude of harmonics 1st and 5th, especially for high rotor
speed. However, fault detection by FFT is not clear at low
speed, and also it is not possible to apply FFT when there are
speed and torque changes. For these cases, a new processing
tool is introduced that is presented next.
III. HILBERT HUANG TRANSFORM ANALYSIS.
Each of the traditional time–frequency analysis techniques
have their own limitations [12, 13]. The consequence is the
misleading energy-frequency distribution for nonlinear and
non-stationary data. The principle of the Hilbert-Huang
transformation is based on the physical time scales that
characterize the oscillations of the phenomena. The local
energy and the instantaneous frequency derived from the IMFs
through the Hilbert transform can give us a full energy-
frequency-time distribution of the data [13]; it would be the
ideal tool for nonlinear and non-stationary data analysis.
A. Empirical Mode Decomposition (EMD).
The EMD method was motivated by computation of
instantaneous frequency defined in terms of Hilbert transform
[13]. As it is well known, for a real-valued signal x(t) the
Hilbert transform is defined by the principal value (PV)
integral.
+∞
1 x(t ' )
y (t ) = P∫ ∂t ' (2)
π −∞ t − t '
where P indicates the Cauchy principal value. This leads to
the definition of an analytic signal, Z(t).
z (t ) = x(t ) + iy (t ) = ae iθ ( t ) (3)
[
a (t ) = x 2 (t ) + y 2 (t ) ]
1/ 2
, θ = arctan( y (t ) / x(t ) )
The instantaneous frequency is then defined by ω=dθ(t)/dt.
In the above process, both the amplitude and instantaneous
frequency are a function of time. One would therefore hope to
construct a time–frequency representation based on Hilbert
transform. In [13] it is shown that for a function to have
physically meaningful instantaneous frequency it is necessary
that the function be symmetric with respect to the local zero
mean, and have the same numbers of zero crossings and
extreme. An intrinsic mode function (IMF) is a function that
satisfies two conditions: first, in the whole data set, the
number of extreme and the number of zero crossings must
either equal or differ at most by one; and second, at any point,
the mean value of the envelope defined by the local maxima
and the envelope defined by the local minima is zero.
The EMD extracts the first IMF by the following sifting
process [12]:
1. Find the upper envelope of x(t) as the cubic spline
interpolated of its local maxima, and the lower envelope,
as the cubic spline interpolated of its local minimum.
2. Compute the envelope mean m(t) as the average of the
upper and lower envelopes.
3. Compute h(t)=x(t) – m(t).
4. If the sifting result h(t) is an IMF, stop. Otherwise, treat
h(t) as the signal and iterate on h(t) through Steps 1–4.
The stopping condition is:
[hk −1 (t ) − hk (t )]2
∑t hk2−1 (t )
< SD (4)
where hk(t) is the sifting result in the kth iteration, and SD is
standard deviation, typically set between 0.2 and 0.3.
The EMD extracts the next IMF (see Fig. 4) by applying the
above procedure to the residue:
r1 (t ) = x(t ) − c1 (t ) (5)
where C1(t) denotes the first IMF. This process is repeated
until the last residue rn(t) has at most one local extreme.
B. Hilbert Huang transform (HHT).
To transform this temporal-space data to time–frequency
space, a Hilbert transform is performed on each IMF
component, and the amplitude and instantaneous frequency for
every IMF at every time-step is computed as shows in the Fig.
5. This result can be projected on the time–frequency–energy
space, with energy defined as the amplitude squared [14].
Fig. 4. Intrinsic mode functions (IMF) for a healthy PMSM. Simulations result
at 1500 rpm.
Fig. 5. Instantaneous frequency for a healthy PMSM. Simulations result at
1500 rpm.
After performing the Hilbert transform on each IMF
component, we can express the data in the following form:
n
x(t ) = ∑ a j (t ) exp⎛⎜ i ∫ ω j (t )∂t ⎞⎟
t
(6)
j =1 ⎝ ∞ − ⎠
One can then represent the signal energy as a joint function
of time and the instantaneous frequency. Such a representation
is called Hilbert spectrum. The residue rn(t) characterizes the
trend of the signal and is treated separately.
The Hilbert Huang transform (HHT) algorithms accurately
analyze physical signals via the following steps:
1. Instantaneous frequencies are calculated based on the
Empirical Mode Decomposition method when intrinsic
mode functions IMFs are generated for complex data.
2. A Hilbert transform converts the local energy and
instantaneous frequency derived from the IMFs to a full
energy-frequency-time distribution of the data.
3. The physical signal is filtered by reconstruction from
selected IMFs.
4. A curve can be fitted to the filtered signal.
IV. EXPERIMENTAL RESULTS
The motors under analysis have been a PMSM of 6000 rpm
nominal speed, 2.3 Nm nominal torque, and 3 poles pair. The
demagnetization analysis for PMSM [1, 4] has carried out
simulation and experimental tests for 6000 rpm, 3000 rpm and
1500 rpm at nominal current and faults of 75% and 50% of
demagnetization. Additionally, speed variations of 500 rpm
around the operational speed have been introduced through the
control. The empirical mode decomposition algorithm (EMD)
and Hilbert-Huang transform (HHT) were implemented by
means of software Hilbert-Huang Transform Data Processing
System (HHT-DPS) [15] from NASA institution. The analysis
of the current is carried out mainly for low speeds, that is to
say, for the cases where the classic methods of fault detection
have not kind characteristic.
The demagnetization is analyzed under different speed
conditions. The stator current is divided in 5 IMF and HHT is
calculated. The results of demagnetization machine are
compared with those obtained from healthy machines.
A. Stable conditions. Constant speed and nominal
torque.
The intrinsic mode functions (IMF) are calculated from the
stator current at speeds of 6000 rpm, 3000 rpm and 1500 rpm
with nominal torque. The difference in magnitude between
healthy and demagnetized machine can be shown with the
instantaneous frequency in the IMF 1 and 2, in the Fig. 6. The
IMF 2 contains the stator current main frequency and the
others correspond to high or low frequencies with regard to the
main frequency. In a similar way to simulations, experimental
results in Fig. 7 and Fig. 8 show the changes in the
instantaneous frequency at 1500 rpm.
In the figures, IMF1 and specially IMF2 show clearly the
fault for the motor damaged. As IMF allow to isolate
frequency ranges in the spectra, they can be used to wrap up
the frequencies that are around those of interest, and to
separate those that are not necessary. For instance, if one IMF
isolates the stator current main frequency, the other IMFs
could be analyzed more easily to increase the precision in the
failure detection., i.e., the analysis can be concentrated in
specify frequency ranges. By this way, the method can be
extended for detection of other electric and mechanics fault.
The Fig. 9 and Fig. 10 show bigger values in HHT for
demagnetization machine at 1500 rpm. Besides, the dynamic
character of the current signal is indicated visually by means
of the magnitude changes around the main frequency of 75
Hz, despite the system is working at a constant torque and
speed.
Fig. 11 shows the instantaneous frequencies at 6000 rpm. The
values of IMF 1 and 2 show once again the demagnetization of
the machine. The above-mentioned comments are easy to
understand taking into account that the demagnetization has a
great influence in the stator current, and the magnetic density
induction is proportional to the speed.
B. Variable conditions. Speed change at nominal torque.
The PMSM usually operates to different speeds and torque
conditions. Under these dynamic conditions, the
demagnetization detection can be carried out by means of
analysis of time – frequency of the stator current. As it could
be seen in the previous section, the proposed method is able to
perform fault detection at a medium speed, constant conditions
Next, it is demonstrated that it is also useful for medium and
high speeds under variable conditions.
Fig. 12 and Fig. 13 depict the instantaneous frequency for
healthy and demagnetized machine when the speed changes
from 1500 to 1000 rpm. The IMF 1 shows the speed change
while IMF 2 is centered on the stator current main frequency.
120
10

100
0

80
Frequency (Hz) -10

60
-20

40
-30

Fig. 6. Instantaneous frequency for a PMSM with 50% demagnetization. 20

-40
Simulation result at 1500 rpm.
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time (S)

Fig. 9. Hilbert Huang Transform (HHT) for a healthy PMSM. Experimental

result at 1500 rpm.
120
10

100 0

-10
80
Frequency (Hz)

-20

60 -30

-40
40
-50

20
-60

-70
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time (S)

Fig. 10. Hilbert Huang Transform (HHT) for a PMSM with 75%
Fig. 7. Instantaneous frequency for a healthy PMSM. Experimental result at demagnetization. Experimental result at 1500 rpm.
1500 rpm.

Fig. 8. Instantaneous frequency for a PMSM with 50% demagnetization. Fig. 11. Instantaneous frequency for a PMSM with 50% demagnetization.
Experimental result at 1500 rpm. Simulations result at 6000 rpm.
 Fig. 14 shows the instantaneous frequency for a
demagnetized machine when a speed change form 6000 rpm
to 5500 rpm occurs. Fig. 15 and Fig. 16 show the resulting
HHT. The differences between healthy and demagnetized
machines are evident, especially at the beginning of the speed
change. The use of the HHT introduces better precision for the
fault detection than others signal processing techniques. The
HHT also shows the condition change and can be used to
implement supervision systems.
Fig 17 shows instantaneous frequencies of IMFs for a
damaged machine with a demagnetization fault. The results
are experimental for a speed change from 1500 rpm to 1000
rpm.
There is a correlation with the simulation results, which
probes the viability of this method for fault detection in
PMSM.
The calculation of the intrinsic mode functions (IMF)
depends on the signal form, the number of the samples and the
current main frequency. Therefore, it has to be carefully
defined, especially when the HHT analysis should be
executed.
The HHT allows to eliminate the undesired frequencies and
concentrates the information in some IMFs, exactly those of
interest for the detection of the fault frequencies.

500
10

400
-10

Frequency (Hz)
-20
300
-30

-40
200
-50

-60
100
-70
Fig. 12. Instantaneous frequency for a healthy PMSM. Speed change from
-80
1500 to 1000 rpm. Simulations result. 0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Time (S)

Fig. 15. Hilbert Huang Transform (HHT) for a healthy PMSM. Speed change
from 6000 to 5500 rpm. Simulations result.

500
10

0
400
-10
Frequency (Hz)

-20
300
-30

-40
200
-50

100 -60

-70

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
Time (S)

Fig. 16. Hilbert Huang Transform (HHT) for a PMSM with 50%
Fig. 13. Instantaneous frequency for a PMSM with 50% demagnetization. demagnetization. Speed change from 6000 to 5500 rpm. Simulations result.
Speed change from 1500 to 1000 rpm. Simulations result.

Fig. 14. Instantaneous frequency for a PMSM with 50% demagnetization. Fig. 17. Instantaneous frequency for a PMSM with 50% demagnetization.
Speed change from 6000 to 5500 rpm. Simulations result. Speed change from 1500 to 1000 rpm. Experimental result.
 V. CONCLUSIONS
In this paper a method for fault detection and diagnosis of
demagnetization is presented, which is based on a newly
developed signal processing technique named as Hilbert-
Huang transform.
The Hilbert-Huang transform is an effective time frequency
analysis method; it can use overalls for the analysis of
nonlinear and non-stationary data.
The demagnetization of a permanent magnet from a PMSM
was simulated by means of finite elements analysis method
(FEA), and results in form of stator current spectrum were
obtained.
The HHT transform allows to analyze the stator current
obtained from simulations for both stable state and dynamic
conditions with speed changes at high, medium and low
velocity. On the whole, their precision and reliability is of
good quality.
The experimental results show also that this method can
effectively diagnose the fault of demagnetization. The results
show an increase in the spectral resolution and also reliability
for the faults diagnosis.
The spectrum HHT is more accurate and easy to interpret. It
does not contain cross frequency and have not problems with
the beginning and the end boundaries. Additionally, the
algorithm HHT is simple and easy to implement in a system
for supervision, fault detection and failure diagnostic by
means of Digital Signal Processor (DSP) with adequate
Development Tools.
The application of HHT allows not only to detect the
failure, but also to know the condition of operation of the
machine and its changes throughout the time.
VI. ACKNOWLEDGMENT
The authors would like to acknowledge the economic
support received from the Spanish Ministry of Science and
Technology for realizing this work under the DPI 2004-03180
Research Project.
Also, the work was supported by the Programme Alban, the
European Union Programme of High Level Scholarships for
Latin America, scholarship No.E04D027632CO, Mr. J.A.
Rosero.
VII. REFERENCES
[1] J. Rosero, J. Cusido, A. Garcia, J. A. Ortega, and L. Romeral, "Broken
bearings and eccentricity fault detection for a permanent magnet
synchronous motor," in The 32nd Annual Conference of the IEEE
Industrial Electronics Society, Paris - FRANCE, 2006, pp. 964-969.
[2] J. A. Rosero, J. A. Ortega, E. Aldabas, and L. Romeral, "Moving
towards a more electric aircraft," Aerospace and Electronic Systems
Magazine, IEEE, vol. 22, pp. 3-9, 2007.
[3] A. A. Arkadan and B. W. Kielgas, "The coupled problem in switched
reluctance motor drive systems during fault conditions," Magnetics,
IEEE Transactions on, vol. 30, pp. 3256-3259, 1994.
[4] J. Rosero, J. Cusido, A. Garcia, J. A. Ortega, and L. Romeral, "Study on
the Permanent Magnet Demagnetization Fault in Permanent Magnet
Synchronous Machines," in The 32nd Annual Conference of the IEEE
Industrial Electronics Society, Paris - FRANCE, 2006, pp. 879-884.
[5] J. Cusido, J. A. Rosero, J. A. Ortega, A. Garcia, and L. Romeral,
"Induction Motor Fault Detection by using Wavelet decomposition on
dq0 components," in IEEE International Symposium on Industrial
Electronics, Montréal, Québec, Canada, 2006, pp. 2406-2411, vol. 3.
[6] Z. Yuping, "Hilbert-Huang Transform and Marginal Spectrum for
Detection of Bearing Localized Defects," in The Sixth World Congress
on Intelligent Control and Automation, WCICA 2006. , Dalian, China,
2006, pp. 5457-5461.
[7] J. Rosero, J. Cusido, A. Garcia, J. A. Ortega, and L. Romeral, "Broken
Bearings Fault Detection for a Permanent Magnet Synchronous Motor
under non-constant working conditions by means of a Joint Time
Frequency Analysis," in IEEE International Symposium on Industrial
Electronics, ISIE 2007, Vigo, Spain, 2007.
[8] R. Yan and R. X. Gao, "Hilbert&ndash;Huang Transform-Based
Vibration Signal Analysis for Machine Health Monitoring,"
Instrumentation and Measurement, IEEE Transactions on, vol. 55, pp.
2320-2329, 2006.
[9] CEDRAT, CAD package for electromagnetic and thermal analysis using
finite elements, User´s guide. Flux 2D. France, 2005, www.cedrat.com
[10] J. F. Gieras and M. Wing, Permanent magnet motor technology : design
and applications, 2nd ed. New York: Marcel Dekker, 2002.
[11] S. Nandi, H. A. Toliyat, and X. Li, "Condition Monitoring and Fault
Diagnosis of Electrical Motors - A Review," Energy Conversion, IEEE
Transactions on, vol. 20, p. 719, 2005.
[12] B. Liu, S. Riemenschneider, and Y. Xu, "Gearbox fault diagnosis using
empirical mode decomposition and Hilbert spectrum," Mechanical
Systems and Signal Processing, vol. 20, pp. 718-734, 2006.
[13] N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H. Shih, Q. Zheng, N.
C. Yen, C. C. Tung, and H. H. Liu, "The empirical mode decomposition
and the Hilbert spectrum for nonlinear and non-stationary time series
analysis," Proceedings of the Royal Society A: Mathematical, Physical
and Engineering Sciences, vol. 454, pp. 903-995, 1998.
[14] E. H. Norden, C. W. Man-Li, R. L. Steven, S. P. S. Samuel, Q.
Wendong, G. Per, and L. F. Kuang, "A confidence limit for the
empirical mode decomposition and Hilbert spectral analysis,"
Proceedings of the Royal Society A: Mathematical, Physical and
Engineering Sciences, vol. 459, pp. 2317-2345, 2003.
[15] N. Huang, "Software Hilbert-Huang Transform Data Processing System
(HHT-DPS)," version 1.4, evaluation ed, NASA GSFC, Ed. United
States: NASA Goddard Space Flight Center (GSFC), 2003-2004.