Boon Logic Amber Analytics Insight¶

Overview¶

Boon Logic's Amber™ platform is the first scalable, autonomously configuring, real-time unsupervised learning product for predictive maintenance and condition monitoring.

Amber™ is powered by the fastest unsupervised Machine Learning (ML) algorithm in the World, the Boon Nano, which quickly and autonomously learns the normal behavior of each individual Asset in your environment. Each Asset-specific model can identify anomalous activity with extreme precision and near 100% accuracy. Amber’s high-level insights allow your operators to know exactly when maintenance is needed, reducing downtime and preventative maintenance costs.

Leverage as an ExoSense™ Transform

The Amber Analytics Insight is an ExoSense™-compatible third-party Insight Module that implements patent-pending unsupervised machine learning methods capable of detecting anomalies from any individual or multi-sensor data stream in real-time.

Capability Highlights

Automatic tuning and training
Individual Asset-level models resulting in multiple analytic outputs
Early detection of anomalous behavior to inform predictive maintenance needs
Scalable and applicable to diverse kinds of signals
Removes reliance on limited human experience to set proper alarm thresholds

Trial Access

This is provided for trial purposes, please contact Exosite for volume pricing

Outputs¶

The following table details the Signals and data generated by the Amber Analytics Insight.

Name Description Type Example Output

Amber Warning (AW)

Representation of Asset "health" based on a threshold of the AM metric auto-determined in Amber (see AM in "Metrics Expanded" section below)

0: Normal, 1: Asset Changing, 2: Asset Critical

The thresholds for the two warning levels are the standard statistical values of 0.95 (outlier) and 0.997 (extreme outlier) respectively

Possible Values: 0, 1, 2

Number (int)

2

Metrics

Additional analytic metrics & metadata (see "Metrics Expanded" section below)

JSON

{
  "state": "Learning",
  "message": "",
  "progress": 1,
  "retryCount": 0,
  "clusterCount": 7,
  "totalInferences": 9,
  "SI": 628,
  "AD": 0,
  "AH": 0,
  "AM": 0.079868,
  "ID": 7
}

Note

The above two outputs are generated as new Signals with their own dedicated data streams. See below for more details on what is included within the Metrics object.

Metrics Expanded¶

Key	Description	Type	Possible Values	Example Value
state	Current state of processing	String	Buffering, Autotuning, Learning, Monitoring	Monitoring
message	When present, additional context about the current state	String	. . .	Reinitializing
progress	Highest percent completion among all possible graduation requirements (see "Configuration Parameters" section below)	Number (int)	0 - 100	61
retryCount	Number of restarts that occurred during Autotuning	Number (int)	0 - 3	1
clusterCount	Number of unique clusters of patterns identified within the model	Number (int)	0 - {Max Clusters}	47
totalInferences	Total number of complete samples (inferences) analyzed in the life of the sensor	Number (int)	0 - infinity	271845
error	Context regarding any processing error that may have occurred	String	. . .	There must be at least 10 patterns for autotuning
SI	Smoothed Anomaly Index, the output of an edge-preserving exponential smoothing filter applied to the raw anomaly indexes of successive patterns With respect to the data seen before, values closer to 0 represent ordinary patterns and values closer to 1000 represent novel - potentially even anomalous - patterns	Number (int)	0 - 1000	382
AD	Anomaly Detection, a binary (boolean) result based on a threshold of the SI metric auto-determined in Amber	Number (int)	0 or 1	0
AH	Anomaly History, a rolling sum of Anomaly Detections over "recent history" (length based on {Samples to Buffer} - see "Configuration Parameters" section below)	Number (int)	0 - {Samples to Buffer}	7
AM	Amber Metric, indicates extent to which an unusually high number of anomlaies have occurred in recent history (based on AH) The values are derived statistically from a Poisson model, with values closer to 0.0 and 1.0 respectively signaling a lower and higher frequency of anomalies than usual	Number (float)	0.0 - 1.0	0.31275
ID	Cluster ID, cluster into which the given input pattern is assigned	Number (int)	0 - {Max Clusters}	16

Tip

Custom Inline Insights can be used to split one or more of the above Metrics into their own dedicated Signal stream(s).

Configuration Parameters¶

The following table summarizes a list of configuration parameters for the Amber Analytics Insight.

Parameter	Description	Single Sensor	Pre-Fused (Fusion)	Example Value
Feature Count	Number of sensor data components included in each sample	Default: 1 Min: 1 Max: 1	Default: 1 Min: 1	8
Streaming Window Size	Indicates the number of successive samples to use in creating overlapping patterns Example: With a Streaming Window Size of 50, each pattern clustered by the Insight is comprised of the most recent 50 samples, meaning each pattern overlaps the previous 49 samples with the latest 1	Default: 50 Min: 1 Max: 500	Default: 1 Min: 1 Max: 1	1
Samples to Buffer	Number of samples to be streamed before Autotuning begins on the buffered samples Rolling window used for Anomaly History (AH) calculation Note: This must be at least 10x {Streaming Window Size}, with a general rule of thumb being to consider 200-500x	Default: 10000 Min: 50	Default: 10000 Min: 50	43200
Learning Rate Numerator	Component of learning graduation requirement for stopping learning See next	Default: 10 Min: 1	Default: 10 Min: 1	1
Learning Rate Denominator	Component of learning graduation requirement for stopping learning The ratio {Learning Rate Numerator} / {Learning Rate Denominator} determines a cluster growth "flatness" threshold During the most recent {Learning Rate Denominator} inferences, if there have been fewer than {Learning Rate Numerator} new clusters created, Learning will stop and Monitoring will begin	Default: 10000 Min: 1	Default: 10000 Min: 1	1000
Learning Max Clusters	Learning graduation requirement for stopping learning During the Learning phase, if the Insight reaches a total of {Learning Max Clusters} in its model, Learning will stop and Monitoring will begin	Default: 10000 Min: 1	Default: 10000 Min: 1	1000
Learning Max Samples	Learning graduation requirement for stopping learning During the Learning phase, if the Insight has processed a total of {Learning Max Samples}, Learning will stop and Monitoring will begin	Default: 1000000 Min: 1	Default: 1000000 Min: 1	250000

Tips

The Samples to Buffer configuration approximately determines which samples are available for the Autotuning state, during which the generalized Min & Max values of any feature(s) will be identified.
- The value to use may change based on the typical operating times and general behavior patterns of the target Asset.
- If historical data is available, consider reviewing the data (e.g. on a Line Chart, Data Table, etc.) to see if there are any temporal (hourly, daily, weekly, etc.) patterns.
- It is also often helpful to talk with someone who works directly with the Asset.
For a Single Sensor stream, given the relationship among sequential time series measurements of a single feature is evaluated, the Streaming Window Size must be greater than 1.
For a Pre-Fused (Fusion) stream, given the relationship among all features included in each sample is evaluated, a Streaming Window Size of 1 is generally sufficient.

Note

Once configuration parameters have been set for a given instance of the insight they cannot be changed. To analyze the stream with different parameters, create a new instance with the desired changes.

Setup Guide¶

This section walks through the setup steps for adding the Amber Analytics Insight to an Asset in your ExoSense application.

Add Amber Analytics Insight to ExoSense Application¶

From your Business Account Home page: Select your ExoSense application, navigate to view the Insight Services added to it, and click "Add Insight Service".

Add Insight Service

This will take you to Exosite's IoT Marketplace where you will find the Amber Analytics Insight among other available Insights. Select that element.

Amber Analytics Insight Exchange Card

From the element page, click the "Add To Solution" button to add this Insight to your application. A window will pop-up with the Terms of Service, check the box and click "Continue".

Terms of Service

This will show a dropdown list of your Murano Solutions with your ExoSense application selected, click "Add To Solution".

Add to Solution

At this point, your ExoSense application is ready to use the Amber Analytics Insight!

Add Amber Analytics Insight to ExoSense Asset¶

Go to the Browse - Group View page and create or select an Asset to modify.

Modify Asset Config

From the Modify Asset page, select a Signal to "Add Transform".

Add Transform

From the Insight Modules dropdown, select "Amber Stream Analytics". From the Function dropdown, select "Single Sensor Stream".

Set the Configuration Parameters to appropriate values for your Asset.

Several new derivative Signals will be created as function outputs. Set names for the Output Signals and click "Save".

Configure Insight

The input Signal you used will now filter toward the top of your Asset's Signal list and the attached Transform will have links to the output Signals it is generating, which will be located near the bottom of the list.

Transform and Outputs Created

Tips

You can now edit the properties of these Signals (e.g. to change the "precision" for the integer-type Amber Warning (AW)).
As noted earlier, you may also consider using Custom Inline Insights to split one or more of the Additional Metrics into their own dedicated Signal stream(s).

Rules and Alerts¶

Select the newly created "Amber Warning (AW)" Signal and "Add Rule".

Add Rule

From the "Standard" Insight Module, select the "Dynamic Threshold" Rule and configure it as desired.

Configure Rule

Dashboard Visualization¶

There are many ways to present data in ExoSense - here is an example showing the Amber Analytics Insight outputs in various ways:

Example Dashboard