AI system transforms large facilities into intelligent environments

January 13, 2017

Max the Magnificent-January 13, 2017

Advanced embedded systems boasting technologies like Artificial Intelligence (AI), artificial neural networks (ANNs) and deep neural networks (DNNs), machine learning, and cognitive (thinking, reasoning) seem to be popping up all over the place -- often targeting application areas that many people would never have considered.

Take the Verdigris AI platform from Verdigris Technologies, Inc., for example. This platform empowers buildings to monitor what is happening inside themselves and to easily communicate with the humans who run them, thereby reducing energy costs and power consumption, and predicting and solving problems before they happen.

The Verdigris solution is far cleverer than you might initially think. When I first read that "Verdigris intelligent sensors learn the individual electrical 'fingerprints' of every single electric device within a building’s walls, down to the smallest iPhone charger," my knee-jerk reaction was to assume that vast numbers of sensors had to be deployed, with individual units associated with every single power socket in a building. This would be incredibly expensive, time-consuming, and complex to implement, so it's fortunate they don’t do it this way.

In fact, the sensors are only associated with the building's main power supply panels. We start with the non-invasive Verdigris smart sensor, which is based on the Hall effect, and which simply clips over one of the feeds coming out of the building's main circuit breaker power panel.

Verdigris smart sensor (Source: Verdigris)

Individual smart sensors are installed non-intrusively on every feed coming out of the panel, with all the sensors being daisy-chained together as illustrated below.

Multiple smart sensors are daisy-chained together (Source: Verdigris)

The smart sensors employ 8KHz digital-to-analog converters (DACs) to sample the current profile with a high-degree of accuracy and resolution. The last sensor in the chain is connected to a wireless data transmitter, which communicates with the cloud. The sensors send hundreds of thousands of data points per second to the cloud-based Verdigris AI, which continuously analyzes and interprets the data, and then communicates its findings to its human supervisors via a web-based interface and/or a mobile app.

Using machine learning techniques based on an artificial neural network, the Verdigris system is trained to recognize the power profiles associated with of a wide variety of devices, from the large motors used in elevators and HVAC systems to tiny smartphone chargers.

When the system is deployed into a real-world facility, all the data from the sensors is constantly uploaded to the Verdigris AI in the cloud. If an unknown power profile is detected, the building's engineers are prompted to identify and label the equipment associated with this profile, and this intelligence is added to the AI's knowledge database for subsequent use in existing and future deployments.

The end result is that, even when a building is fully occupied and its power systems are highly loaded, it's still possible for the Verdigris system to tease out the fact that someone on the third floor just plugged in an electric shaver, for example.

The Verdigris Tracker is a mobile web app that provides real-time event tracking and notifications to the facility's engineering team. If the Verdigris AI detects any abnormalities or building drift, it will use the tracker to inform the building's engineers.

Verdigris Tracker (Source: Verdigris)

If monitoring power consumption and identifying which devices are active at any particular time were all the Verdigris system could do, it would still be a valuable tool, but it goes much farther than this. Using a physics-based modelling approach, the AI is also trained to recognize the power profiles associated with various failure modes associated with different pieces of equipment. This allows it to detect and identify units that are causing spikes in energy usage, that are potentially going to fail, or that have already failed, and to immediately alert the associated maintenance team.

Automated fault detection with notifications that can prevent
unexpected equipment breakdown (Source: Verdigris)

To date, Verdigris has saved millions of dollars in reduced energy costs and equipment breakdowns for customers like Jabil, Honeywell, Hyatt, Starwood, Marriott, and even NASA, which is researching ways to apply Verdigris technology to predict failure on potential future space missions.

An example of the real-time interactive lobby dashboard that's deployed at the
new Jabil Blue Sky facility in San Jose, CA (Source: Verdigris)

The Verdigris AI becomes increasingly smart and more connected as existing Verdigris customers use the system and as more facilities come online. The end result is to provide the AI with a massive repository of incredibly-specific, granular information about buildings, their energy use, the way in which their internal systems and appliances work, and the signals these systems send when they start to degrade and fail.

At this time, Verdigris are targeting their platform for use in large buildings and facilities, but I wouldn’t be at all surprised to see residential equivalents being deployed in the not-so-distant future -- especially in new home construction -- as part of the migration to smarter, more efficient abodes. What do you think about all of this?

Loading comments...

Parts Search

Sponsored Blogs