Skip to content

Simple Anomaly node library

Introduction to simple anomaly nodes

The anomaly nodes help to detect anomalies in a data set.

The simple anomaly nodes in Visual Flow Creator are of two types:

  • Training model
  • Reasoning model

Note

Region deviation

Support for "Simple Anomaly nodes" is in progress for Region Europe 2.

Terms and definitions

Cluster: A cluster is a high density region in a defined space. Each cluster is a set of data points which defines a dense region.

Anomaly points: The number of anomaly points available outside a data cluster.

Neighborhood: The neighborhood of a point is a set of all points that are within a distance set by the neighborhood parameter.

Epsilon (ε or eps): The epsilon (ε) defines the distance defined of neighborhood around a certain point in the selected algorithm (Euclidean, Manhattan, Chebychev).

Anomaly extent: Distance between an anomaly point and its centre point (ε) of the cluster. Anomalies are minimal for lesser anomaly extents. For greater distances, anomalies are larger.

Algorithms: Euclidean: The Euclidean distance is defined by a straight line between two points. Manhattan: The Manhattan distance measure algorithm finds its application on high dimensional vectors. The sum of the absolute differences among their coordinates is defined as the distance between two points. Chebychev: The maximum distance from the centre of a cell to its adjacent cells centre is defined as the Chebychev distance between two points.

Training model node

Training model

The training model node defines a new model of some data sets in Visual Flow Creator.

The data set information gets configured as read time series node with one condition - you can select only one aspect. The data set is defined by asset/ aspect/ variable. The data set can be selected using "Select asset aspect" selection dialog box.

Training model created will be stored in MindSphere exchange storage.

Training model node properties

You can confirgure the node by editing its node properties.

Training model node

Field Name Description Mandatory
Name Select a display name for the node. No
Model Name Name the new training model. Yes
Description Enter the description. No
Topic Import the required time series data from the "Topic" field. The menu will Yes
redirect to select read time series data with
asset /aspect variable.
Topic Summary Each asset will have aspects which are defined by variables. Select the required Yes
variable of the selected aspect of its asset to import the data in the
"Training model" node properties. This will autofill the "Topic Summary"
Mode Select the mode from the drop-down menu. "Mode" is either defined by period of Yes
time or an interval.
Period
Period Select the time period from the drop-down menu. The time period drop-down has Yes
values defined in minutes, hours, days and weeks.
Offset Select the offset value for balancing the effects. The offset values are Yes
defined in seconds.
Interval
Timezone Select the required time zone from the drop-down menu. Yes
From Select the start date and time from the calendar and time menu. Yes
To Select the end date and time from the calendar and time menu. Yes
Epsilon Set the ε for the new model. Yes
Min. points per cluster Define the minimum number of points which are in each cluster. Yes
Distance Measure Algorithm Select the required distance measure algorithm from the drop-down. Yes
Three algorithms are available:
- Euclidean
- Manhattan
- Chebychev

When you set the topic in "Select Asset/ Aspect" dialog box, select the necessary asset, aspect and its variable as shown below:

Asset Aspect

Jobs run involved

While injecting a timestamp to the "training model" node, a set of batch jobs run in the following series:

  • Fetching IoT data
  • Create model job
  • Export job and generating output

Reasoning model node

The reasoning model node is used for analyzing data for a stored training model.

Reasoning model

Reasoning model node properties

You can confirgure the node by editing its node properties.

Reasoning model node

Field Name Description Mandatory
Name Select a display name for the node. No
Model Name Select a training model from the menu. Yes
Description Enter the description. No
Topic Import the required time series data from the "Topic" field. The menu will Yes
redirect to select read time series data with asset /aspect /variable.
The import will capture the actual details of the model selected. However,
the aspects and variables imported should be same as that selected for
the "Model Name".
Topic Summary Each asset will have aspects which are defined by variables. Select the required Yes
variable of the selected aspect of its asset to import the data in the
"Training model" node properties. This will autofill the "Topic Summary" field.
Mode Select the mode from the drop-down menu. "Mode" is either defined by period of Yes
time or an interval.
Period
Period Select the time period from the drop-down menu. The time period drop-down has Yes
values defined in minutes, hours, days and weeks.
Offset Select the offset value for balancing the effects. The offset values are defined Yes
in seconds.
Interval
Timezone Select the required time zone from the drop-down menu. Yes
From Select the start date and time from the calendar and time menu. Yes
To Select the end date and time from the calendar and time menu. Yes

When you select a model name, the data of the selected model will be imported to the reasoning model node for analysis.

All models

The property details of the selected model will also be displayed to the user.

Jobs run involved

While injecting a timestamp to the "reasoning model" node, a set of batch jobs run in the following series:

  • IoT import job creation
  • Import the job model
  • Apply the job model
  • Export the job and save in the database

Using simple anomaly nodes

Example scenario

Configure a training model node. Analyze the anomalies in the model.

Objective

To analyze the anomalies in the newly created training model with the help of a reasonaing model node.

Requirements

Import the data from Fleet Manager to create a training model.

Configuration of the traning model is given below:

Field Name Data
Name CPU temp measure
Model Name Temperature Model
Description epsilon 0.5
points 5
Topic 10368277c5c94d3b9659de28ee1ea85a
Topic Summary Asset: 10368277c5c94d3b9659de28ee1ea85a (GPU)
Aspect: CoreTemperature
Variable: Double
Mode Interval
Period
Timezone UTC
From 2019/01/01 01:00:00
To 2019/03/12 09:59:00
Epsilon 0.5
Min. points per cluster 5
Distance Measure Algorithm Chebychev

Procedure

  1. Select a reasoning model node from the "simple anomaly" nodes section.
  2. Configure the reasoning model node. The configuration of the node is given below:

    Field Name Data
    Name CPU temp measure reasoning model
    Model Name Temperature Model
    Topic a1ac2fa99bd94d95bef997e413eaab70
    Topic Summary Asset: a1ac2fa99bd94d95bef997e413eaab70 (CPU)
    Aspect: CoreTemperature
    Variable: Double
    Mode Period
    Period
    Period 1 hour
    Offset 2 seconds
  3. Insert an input timestamp and a mesage payload to the reasoning model node to generate the output in the message payload.

Timestamp

Note

The "Topic Summary" properties in training model node should match the "Topic Summary" properties of the reasoning model node. This means that aspects and the variables for both should be same else the output will conflict and result in a failure.

Result

Inject the timestamp to process the flow.

You get the following result in the message payload:

Debug

Some of the initial anomaly points are highlighted in the graph as well in the array description.

Any questions left?

Ask the community


Except where otherwise noted, content on this site is licensed under the MindSphere Development License Agreement.


Last update: April 29, 2022