Advantech WISE-750, Руководство пользователя

Страница: 32 / 44

WISE-750 User Manual 26
2. Enter the file count in the box.
3. Click "…" button to enter a destination of your data.
4. Click on the Start button then the data logging will be started.
4.7.2 Feature Extraction Mode
The Feature Extraction Mode allows you to set the alarm using the feature values we
provided already in WISE-750. They are maximum, minimum, peak value, peak to
peak and RMS respectively. To employ this mode, the alarm type of WISE-750 needs
to be set to "Feature Extraction" in Configuration Page.
Alarm Threshold Setting
You can choose which feature value you are concerned about by checking the box
and then entering the criteria. Finally, press the button "Save To Device" to set the
alarm criteria to WISE-750. After that, the condition monitoring using criteria setting
will be started.
For the details of all the feature values, please refer to Appendix B.
4.7.3 Intelligent Mode
Machine Learning Model Building
The Intelligent Mode operates the condition monitoring by machine learning algo -
rithm. The WISE-750 allows you to set models for each channel. The training data
set is pre-processed by FFT to transfer from time domain to frequency domain. Fol -
low the steps below to employ Intelligent Mode in WISE-750:
1. Set the alarm type to Machine Learning and the channels in Configuration page.
2. Switch to ML Modeling page
3. Set the number of data set you'd like to train the model with. Every data set is
saved to a file. Each file contains data count of half of the section length. For
example, if you set the file count to 100 and section length to 2048, there will be
100 files and each file containing 1024 data comes after FFT (Fast Fourier
Transform).
4. Set the destination. This would be the path of the data set. We strongly suggest
you to fix this path once you decide to use this mode.
5. Press start then the datalogger will start to retrieve the data set for machine
learning. The datalogger will follow the parameter you set in Configuration to get
the data. If you set the trigger, the datalogger will get data until the trigger hap -
pens.
6. After retrieving the data set, the dataset will be listed in the column Dataset
Selection. Select the dataset you'd like to train the model with.
7. Press start. This will train the model and also download the model to each
selected channel.
Alarm Threshold Setting
The training session will build and download the model to the WISE-750. In the infer -
ence session, the WISE-750 will indicates if the iteration is OK or NG. The WISE-750
will give a score to indicate the status of the measured machine/component. The
score is calculated using success rate in the past N iterations, while N is user-config -
urable in Alarm Setting page. For example, if the number is set to 200 and there are
160 Oks in the past 200 iterations, the score will be 160/200 * 100 = 80. Users need
to set an alarm threshold for the score. Hence, the WISE-750 will generate the alarm
when the score is lower than the threshold. Then, press the button to set the parame -
ters for the device.