Enhancement of condition monitoring information from the

control data of electrical motors based on machine learning

techniques

Sensors(Data aquazation)

preprocessing

Feature Extraction

Feature selection

Diagnosis/prognosis

Sensors: Therrmal Monitoring-Infrared camera, Thermistors.

Vibration Monitor-Vibration Transducers

Acoustic Emission-Wide band or resonant piezo electric sensors

Electrial Signature Analysis-MCSA sensor (Motor current signature analysis)

Pre-processing: Gear boxes, Centrifugalpumps, Reciprocating compressors - CWT Analysis

Bearing Faults -- Frequency Spectral Subtraction based on DWT and SWT

Centrifugal Pump -- MSB technique for extracting non-linear characteristic of

 current signal from the motor driving a centrifugal pump systems under different

impeller faults.

Feature Extraction:

Time Domain Analysis:

Root Mean square

Kurtosis Feature extraction of raw vibration signal collection of time

Crest Factor indexed data collected representing in acceleration ,velocity

Or Proximity based on the type of transducer used to

Collect the signals. Vibration signal are collected from rot

-ating machine using vibration transducers in time domain.

Frequency Domain Analysis:

Discrete fast fourier Transform (DFFT)----In FDA techniques are used to know the information based

On frequency characteristics that are not easily observed in

In time domain. Each sine wave will be presented in spectral

Component.each component of a rotating machine produces a single frequency. However,we do not

often see these produced frequencies individually in the measured signal: we see a summation of the
signals that the sensor measured.The spectrum of the frequency components generated from the time
domain waveforms makes it easier to see each source of vibration.
 Wavelet Transformthe raw data as a function of time, also

most widely used for analysising and assessing the acquired signals ,it is limited
Time-Frequency Domain: when analysing the non-stationary and non-linear signals for fault diagnosis. Data-
driven .Techniques such as EMD and ITD allow multi- components of signal to be
Short-time fourier transform
decomposed into different simple intrinsic components where the effective
features can be extracted and used for detecting and diagnosing the faults.

Envelope analysis: STFT: STFT, which computes the DFT by decomposing a signal into shorter
segments of equal length using a time-localised window function.
Empirical mode Decomposition
Wavelet: Wavelet analysis decomposes the signal based on a family of ‘wavelets’.
Intrinsic Time-scale Decomposition

Feature Selection:

Diagnosis:

SVM

Equipment:
Sinocera Ye6232B multi channel for acquiring the experimental data.

Software:YE7600 softwarre where the vibration levels and current

signals could be monitoried and analysed in real time.
 Developing Anomaly detection, Diagnostics and prognostics

for condition monitoring with limited historical

data in new application Such a tidal power

Vibration Data

preprocessing

Feature Extraction

Classification

Diagnosis/prognosis

Sensor:

HS1000- High resolution accelerometer Data-DXI accelerometer

Sensor for vibration.

Dta MIniing: To discover trends and relationship b/w parameters with
in low resolution datasets from the HS1000 tidal turbine.

For data mining CRISP-DM process model utilized.

Data Analysis:

Correlation : Discover relationships b/w pairs of variables.

Principal Component Analysis- It is used to reveal the complex

relationship by transforming data into projections along different axis.

Change point analysis: Data parameters change notably overtime used

to define in time series of data or detect changes in data missed by
visual inspection.

Anomaly Detection Techniques:

Probabilistic—GMM,KDE

Distance based—K Means clustering, K Nearest neighbors

Domain Based-One class SVM, Envelop Fitting

Reconstruction Based-curve fitting

Pre-processing:
--Order Tracking

Computed resampling

Vold Kalman filter

Enveloping
Feature Extraction:

Time Domain

RMS

Crest factor

Kurtosis

Skewness

Frequency Domain:

FFT

Time-frequency Domain:

STFT

Wavelet Transform

Cepstrum analysis

Classification:
SVM
Decision Tree
KNN
Feature based diagnostics process

Vibration Data

Vold Kalman filter

RMS

Z-score

Classification(SVM,Decisi
on Tree,KNN)

Prognosis:

Prognosis Method:

-Time To Failure Analysis

- Stressor based

- Degradation based prognostics

Software: Matlab
Wireless sensor Network Platform for Railway Condition
Monitoring

Sensors: MEMS sensors-MEMS accelerometer and geophone.

ADXL 335 Accelerometer sensor.
Pre-processing circuit: NI-3226 sensor node
NI-9791 real time gateway
Feature Extraction:
Frequency analysis- Raw signals analysis is identify like
minimum magnitude and maximum magnitude, peaks,
threshold level.
Frequency analysis of a sinusoidal signal on matlab.
Frequency analysis of the signal for various sampling intervals
in LABVIEW.
Diagnosis:
-Tri-axial acceleration magnitude signal
MATLAB
-Vertical acceleration signal and FFT
Equipment Used:
-MEMS Sensors
-Arduino Liquid,NI-3220,NI-9791
- R-2R ladder CKT
Software: -Labview
-Matlab
- NI-max

Accelerometer sensor

NI-3226 sensor node

NI-9791 real time

gateway

Frequency Analysis

Tri-axial and vertical

acceleration signal and
FFT
Intelligent-Methods for condition monitoring rolling bearings
Using Vibration data

Sensors :
Vibration data from Torsions Sensors-Torque Transducer and
Encoder.
Pre-processing:
Frequency domain:
Compressive sampling for time -frequency
representation using FFT
Feature Extraction:
Time Domain Analysis-
Statistical functions
Time synchronous averaging
Time series regressive models
Filter-based models
Stochastic Parameter techniques
Blind source separation
Frequency Domain Analysis-
DFT
FFT
Frequency spectrum statistical feature
Time-Frequency Domain Analysis-
Wavelet analysis
Short time fourier Transform(STFT)
HHT
EMD
Nigner-vile Distribution
Local Mean Decompoisition
Spectral Kurtosis and Kurtogrsm
Linear subspace Learning-PCA
ICA
LDA
Feature selection:
Filter model-based
Wrapper Model Based
Embedded Model-Based
Diagnosis:
SVM
Health condition on bearings
LRC
ANN

CSFR framework assessment and validation

CS is used to learning the high dimensional data
And combining the efficiency of CS in improving the rolling
bearings condition monitoring.
Methods include-----CSFR framework based methods
CSLSL based methods
CS-SAE-DNN methods

Designing new methods for vibration

based bearing for condition monitoring
Vibration Data

Frequency Domain

Time-frequency
domain( NVD,EMD,HHT,
Linear subspace
Learning)

Embedded Model-
based

SVM,LRC,ANN and CSFR

Hidden Markov Model Supported Machine Learning for
Condition monitoring of DC-link capacitors

Currents measurements

Frequency Domain

FFT

FFT smoothing

SVM ANN

Lamada Lamada
Type equation here .
SVM ANN

SVM,LRC,ANN and CSFR