Real Machine Learning for Asset Health Monitoring

Amber detects anomalies in complex assets and environments by autonomously training for the normal instead of the abnormal.

5 minutes

Time to create a model from scratch

6 weeks

Average warning before asset failure

Any user

No data science skills required to build ML models

Any platform

Plugin for any IoT platform

Enhance reliability with AI-driven insights

Make smarter decisions with complete visibility into the health of your operations. Amber detects equipment noncompliance earlier than any other predictive analytics tool, giving you more time to schedule maintenance, order parts, or change production settings, all with data from your existing historian.

Accurate prediction

Detect early signs of failure as soon as they occur without annoying false alarms

Easy to use

ML designed to be configured by reliability engineers and plant staff

Quick to scale

Deploy thousands of models in weeks

Complete visibility into your equipment health

Finding the abnormal is easy once you understand normal. Even with the most complex equipment, Amber trains from data collected during normal operations, instead of relying on failure modes and programmed logic. Starting from normal is the only way to have insights you can truly trust.

 

Amber’s Compliance Score is based on an individualized, high-dimensional, unsupervised machine learning model of your asset. The score indicates ongoing asset health and deviations from normal operating behavior.

 

Normal: 51% - 100%


Changing: 26% - 50%


Critical: 0% - 25%

Insight into the source of the problem

Feature Significance sheds light on which features are the culprits in causing a drop in Compliance Score. Being able to swiftly pinpoint the origin of the problem allows for prompt and cost-effective issue resolution.

All in your existing platform

Three steps to deployment

Configure

To configure Amber, choose 1– 500 tags to be incorporated into a model. The selected tags should be related to the health of the equipment or process. Amber will learn the relationships that exist between the tags during training.

 

Types of tags to include:

  • Production inputs and outputs
  • Equipment settings
  • On/off
  • Temperature
  • Current
  • Flow
  • Speed
  • Pressure
  • Displacement measurements

Train

Amber’s auto-learning algorithm trains itself, enabling deployment in minutes, not weeks. During training, Amber learns hundreds of relationships between the tags selected during configuration. As training progresses, Amber’s learning curve starts to level, indicating that Amber is becoming familiar with the asset’s normal operations. Once the learning curve plateaus, training is complete.

 

  • Trains a unique model from scratch for each asset
  • No programming required
  • No data scientist required

Monitor

After training concludes, Amber seamlessly shifts into monitoring mode. Throughout this phase, streaming data is assessed against the model established during training. For every sample, a Compliance Score and Feature Significance are generated. Samples closely resembling the model yield higher Compliance Scores, whereas significantly deviating samples result in lower Compliance Scores.

Connect your existing platform

Programming interfaces

Messaging protocols

Amber can be integrated into your existing data management platform for deployment entirely on-premise or in the cloud. All of Amber’s analytics can be visualized in your existing user interface (PowerBI, PI Vision, VantagePoint, etc).

Our supported platforms

Native integration with PI Web API

Native integration with Ignition REST API

Native integration with Azure Event Hub

Native integration with Cumulocity

Data-driven decision making

From December 11th to January 6th, Amber trained using live data from the compressor. From January 29th to February 11th, the Compliance Score decreased, eventually falling below 50% and indicating that the compressor was in a new never-before seen state.
On March 9th, the Compliance Scored had decreased to 24%, signifying that the problem with the compressor was critical. Staff members onsite were notified of the score and the associated Feature Significance values, which pointed to Stage 3 of the compressor. A work order was created for maintenance.
Two maintenance technicians inspected Stage 3 of the compressor. No obvious problems were noted during an initial overview, but upon closer inspection the technicians noticed a few small cracks on the cooling bundle. After further investigation, several tears were found on the cooing fin. Within 72 hours the repair was complete and the unit was back online.

Make failure and downtime an anomaly