Icon

Predictive_​Maintenance_​Model_​Training

<p>Predictive Maintenance<br>Anomaly Detection. Time Series AR Training<br><br>This workflow trains an auto-regressive model for anomaly detection:<br>- Filter the data to training data covering only normal functioning<br>- Loop over each frequency column at a time<br>- Train an auto-regressive model using 10 past values as predictors<br>- Calculate in-sample prediction error statistics<br>- Save the model and prediction error statistics for deployment</p>

Predictive Maintenance - Training

This workflow trains an auto-regressive model for anomaly detection

This workflow trains an auto-regressive model on data describing a functioning rotor for anomaly detection. The input data includes 313 time series of spectral amplitudes from 28 sensors located on a rotor.

Model training
read AlignedData.csvproduced byTime Alignment & Visualization313 time series
CSV Reader
save errorstats
CSV Writer
Lag = 10 10 samples as past history
Lag Column
loop on all 313 columns i.e. on all frequency bins on all sensors
Column List Loop Start
Loop End
train a Linear Regression Model for each time series column
Linear Regression Learner
using Last Value for missing values
Missing Value
ex:[500-600]+Amp -> Amp
Column Name Replacer
Save model
Jan-Aug 2007 is the training set
extract Jan-Aug 2007
calculate error stats
Error Stats
Regression Predictor

Nodes

Extensions

Links