Skip to main content
1
Main Keywords
Industrial IoT
• Internet of things (IoT) is the network of physical
devices embedded with sensors and connectivity which enables
these objects to connect and exchange data.
• Industrial IoT is IoT in the context of industrial processes such as
manufacturing.
Machine Learning
• A program or system that builds (trains) a predictive model from
input data.
• The system uses the learned model to make useful predictions
from new (never-before-seen) data drawn from the same
distribution as the one used to train the model.
• Supervised Learning and Unsupervised Learning
Anomaly Detection
• Anomalies are unusual patterns that do not conform to the
expected behaviour of a system
• Anomaly Detection is a technique used to identify anomalies
Use Case - Introduction
• Detect anomalous start-up sequence of a
hydro electric power generator.
• Types of Anomalies
• Value
• Behavioural
• Parameters
• Power generated
• Generator RPM
• Generator Cooling Water Temperature
• Ambient Temperature
• Various bearing Temperatures
Normal vs. Anomaly
Solution Overview
• Artificial Neural Network model is trained on normal generator startup sequence (best fit)
• This model is then applied to new generator sequence to detect anomalous behavior
• Anomaly is detected by reviewing the reconstruction error (predicted vs actual value) and applying thresholds derived during model
training
NEURAL NET
(MLP)
PREDICT
INPUT DATA STREAM
(Generator Startup Sequence)
TRAINING DATA SET
(Good Startup Sequence)
ANOMALY
Minimize Loss
Function
ERROR > α
α = Error Threshold
Model Building - Methodology
Input
Raw Data
(7 Parameters)
+
Engineered
Features
(Features built on
8 parameters)
Output
Prediction of
Outcome
parameter
Calculate
Root Mean
Square Error
(RMSE) for
the model
Compare the
RMSE(model)
with
RMSE(new
sequence) to
detect
anomalies
Model Performance - Actual vs. Predicted
Training Normal Sequences
RMSE = 0.01317
Testing Anomalous Sequences
RMSE = 0.302729
Anomaly Detection - Modes
Batch Mode
• Complete start-up sequence is
passed as input.
• Compare RMSE(model) to
RMSE(sequence)
Scenario RMSE
Normal 0.012556
Anomalous 0.302729
RMSE of Model = 0.01317
Anomaly Detection - Modes
Real Time Mode
• Pass streaming data from the
sensor as input.
• Anomaly if RMSE is increasing over
time.
Summary
• To identify anomalies, identify the normal.
• Create features that describes anomalies.
• Simulate the normal behaviour with a machine learning model.
• Compare the predictions with actual.
• Deploy in batch mode or real time mode.
Questions
13