# Failure Prognosis for Permanent Magnet AC Drives Based on Wavelet Analysis Wesley G Zanardelli Elias G Strangas and Selin Aviyente zanardel egr msu edu strangas egr msu edu aviyente egr msu edu

W
Shared by:
Categories
-
Stats
views:
4
posted:
12/18/2009
language:
English
pages:
1
Document Sample

```							                                                                                                                                                                        Failure Prognosis for Permanent
Magnet AC Drives Based on
Wavelet Analysis
Wesley G. Zanardelli, Elias G. Strangas, and Selin Aviyente
zanardel@egr.msu.edu                         strangas@egr.msu.edu                               aviyente@egr.msu.edu

•Objectives                                                                                                                                                                         •Linear Discriminant Analysis                                                                               •Linear discriminant analysis applied to the 64
•Detection of non-catastrophic faults in Permanent                                                                                                                                    Dk (x ) = x1 1k + x2 2k + , , + x N Nk + N +1,k                           k = 1,2 ,...,K                 samples beginning 8 samples prior to where detection
Magnet AC machines which lead to reduced life and                                                                                                                                    •Categorization of xi into j                                                                              occurred
eventually failure                                                                                                                                                                     D j (x ) > Dk (x )         for every k ≠ j
•Stator Faults                                                                                                                                                                       •Iterative training procedure for the weighting                                                     •Experimental Setup
•Insulation failures                                                                                                                                                               coefficients makes adjustments to j and l following an
•Resistance changes                                                                                                                                                                initial guess, where
Dl (x ) = max l ≠ j [D1 (x ),           , DK (x )]
j   (i + 1) =   j   (i ) + axi               l   (i + 1) = l (i ) − axi
•Discrete Wavelet Transform (DWT)                                                                                                                                                     •and a is a gain constant
•Wavelets have finite energy concentrated around a
point which helps to localize irregularities in a signal
•Can give a sparse representation of a fault                                                                                                                                       •Faults Explored
•Can choose different basis functions (mother wavelets)                                                                                                                              •Series Resistance (5 and 10ms)                                                                     •Typical Results
Iq for Series Resistance Fault (10ms)                        Iq for Turn-Phase Short (10ms)
to achieve the best results for a specific application                                                                                                                                 •Intermittent increased series contact resistance                                                                                   28                                                         29

Current (A)

Current (A)
•Coefficients can be realized using a filter bank                                                                                                                                      •A normally closed switch and a resistance in parallel                                                                              26
24                                                         28
22

•FIR filter coefficients h1 and h0 based on the scaling                                                                                                                                are added in series with one of the motor phases                                                                                    20
18
27

and wavelet functions                                                                                                                                                                                                     Inverter         Electronic
Switch
PMAC
Motor
6
UDWT for Series Resistance (10ms)
6
UDW T for Turn-Phase Short (10ms)

5                                                           5

Scale

Scale
h1 ( −n )                      2             dj                                                                                                                                            R                                                                           4
3
4
3

c j +1                                                                            h1 ( −n )                  2            d j −1
•Turn-to-Phase Short (5 and 10ms)                                                                                                 2
1
2
1

h0 ( − n )                     2             cj                                                                                        •Insulation failure in the stator windings of the motor                                                                         4
Classification
4
Classification

h0 ( − n )                 2            c j −1                                •A normally open switch is added between a                                                                                      3                                                           3
2                                                           2
winding and its corresponding phase                                                                                             1                                                           1
0                                                           0
2.1      2.2       2.3     2.4                            2.1      2.2     2.3      2.4
Time (s)                                                  Time (s)

•Undecimated Discrete Wavelet Transform                                                                                                                                                                                                                                                   •Algorithm Performance
(UDWT)                                                                                                                                                                              •Analysis Methods
•Shift-invariant representation of the DWT                                                                                                                                           •Field oriented currents are used since the fundamental                                                                                                        Number of False
Fault Inception      Fault Clearing
Test Description                                                          Total / Detected /   Total / Detected /
•Realized using the “Algorithme à Trous”, which omits                                                                                                                                electrical frequency is not present                                                                                                                            Detections
Classified Correctly Classified Correctly

downsampling and inserts zeros between filter                                                                                                                                        •UDWT applied to measured q-axis current                                                             Healthy                                                           0              0/0/0                0/0/0
Series Resistance (5ms)                                           0              2/2/2                2/2/2
coefficients at each successive scale                                                                                                                                                •Daubechies D4 wavelet used                                                                          Series Resistance (10ms)                                          0              2/2/2                2/2/2
•Decomposition performed for 6 scales                                                                Turn-to-Phase Short (5ms)                                         0              2/1/1                2/2/2
Original Signal                                      DWT of Original Signal                                   UDW T of Original Signal
Turn-to-Phase Short (10ms)                                        0              2/2/2                2/2/2
28
26
4                                                        4                                                  •Inception and clearing of faults are identified separately
Amplitude

24                                                        3                                                        3
Scale

Scale

22                                                        2                                                        2                                                  •Detection Algorithm
20
18
Original Signal
1
DWT of Original Signal
1
UDW T of Original Signal
•A threshold is applied to the weighted energy of the                                            •Conclusions
(S hifted by 6 Samples)                                  (Shifted by 6 Samples )                                  (Shifted by 6 S amples)
28
26
4                                                        4                                                     UDWT at each time instant                                                                          •Detection and classification of machine faults which
Amplitude

24                                                        3                                                        3
•Threshold is set to 40% greater than the largest
Scale

Scale

22
20
2                                                        2                                                                                                                                                        manifest themselves in the stator current is achieved
18
100     120      140             160
1
100        120      140               160
1
100      120      140            160
observed on healthy motors                                                                         •Data from an exhaustive set of operating conditions is
Sample Number                                               Sample Number                                          Sample Number
•Classification Algorithm                                                                             necessary to develop a robust algorithm

```
Related docs
Other docs by peisty474