MORE

LPDG Anomaly
DETECTION

we inspire + we innovate + we enable

Anomaly Detection in large data sets builds the basis for a variety of different use cases.

The detection of anomalies is today not only important for the implementation of use cases such as Predictive Maintenance with data from IoT devices and sensors, but also for an early detection of intrusions into computer networks in the field of Cyber Security, or even the detection of deviating data sets or extreme states in other large data sets like huge amounts of Service Orders.

In contrast to rule-based approaches, the LPDG anomaly detection uses methods from the field of machine learning to detect even previously unknown anomalies in the data.

THE CHALLENGE

Anomalies has to be detected in large amounts of IoT data (> 300.000 devices) considering the "device history" represented by messages of the past six months. The analytical application has to learn from the device history and detect also unknown anomalies.

The extraction of the raw-events sent by the devices and the data-preparation must be implemented in a highly scalable way in order to handle the future increasing number of devices.

For an informative reporting, the model assigns every device to a cluster based on the current and historical device status represented by the events that were sent by the device.
The clusters for detailed reporting are grouped into categories that enables the focus only on devices with a specific status.

THE SOLUTION

Implementation of machine-learning algorithms (unsupervised learning with K-Means-Clustering) on the LPDG Analytic Platform to cluster mass data for detection of different kinds of anomalies in the sensor data.
Usage of Big Data technology to scale with the increasing number of devices and event data. Visualization of the results on different levels of details including a “traffic light system”

Clustering of the devices into detailed clusters representing the device history
Detailed analysis by enrichment of the clustered devices with information about the location and firmware-version
Flexible analysis by using filtering and drill-down even to device level
Notification in case that the number of devices in spescif clusters significantly increases

THE VALUE

Detection of various categories of anomalies from the current and historical status of the devices.
Representation of the "device history" by the assignment to the different clusters that takes the past six moths into account and not only the current status of the device.

Reporting and summarization on different levels like status, location, firmware version or down to device level.
Integration of the Insights on device level into marketing, repair and maintenance processes

LPDG Anomaly Detection based on machine learning makes it possible to gain completely new insights from mass data that would not be discoverable with traditional analysis methods

"Big Data and Data Science lift the value of data ensures competitiveness, decision-making and digital transformation."