The growing number of sensors on industrial equipment is creating what might be called “fast data,” a flow of information that needs to be analyzed, often on the fly. Companies on the forefront of the Industrial Internet say analytics systems can squeeze tangible benefits from a flow of machine data.
Energy management software company EnerNoc has been handing data from utility meters, its controllers and buildings management systems for a decade. Earlier this month, it said it started using predictive analytics software from artificial intelligence startup Numenta, part of an effort to use cutting-edge analytics tools to increase revenue.
Called Grok, Numenta’s software analyzes data from utility meters to predict how reliably an EnerNoc customer will participate in a request to lower energy use in two hours. EnerNoc provides demand-response services where it curtails electricity at commercial customers, such as factories or big retail stores, for utilities. Utilities rely on reducing peak energy demand through the services of companies like EnerNoc.
Rather than store data in a database, Grok automatically creates a data model that identifies significant attributes in a raw data feed to make predictions. For example, after analyzing meter data over several hours, it can forecast energy usage patterns in a building or spot a fraudulent transaction. Another Numenta customer uses Grok to be alerted when potential mechanical problems can occur in offshore wind turbines, a way to reduce costly maintenance trips.
The Challenge of Industrial Data
The software from Numenta, based in Redwood City, Calif., is an example of the innovation that’s occurring to handle streams of information from industrial equipment, sensors, and other devices not traditionally considered computers. Experts say analyzing data from instrumented machines is becoming one of the most challenging aspects of big data because the speed at which data is produced and the wide range of data formats from industrial equipment.
In EnerNoc’s case, its software has to respond to a grid operator’s request to lower electricity use and send a signal to control equipment in less than a second. EnerNoc also has algorithms to analyze data that comes in every five minutes, such as a lighting control system and performance data on a building’s heating and cooling system.
Even if it’s collected less frequently, the data can still be time sensitive. A building owner may want to dim lights or temperatures in an office building to save energy and needs to know right away if night settings haven’t gone into effect. “The time stamp and the frequency of the data runs the spectrum,” says Gregg Dixon, EnerNoc’s senior vice president of marketing and sales. “It’s really application-specific.”
EnerNoc’s core software, which processes gigabytes of data every day, is built on a utility-industry version of Oracle and a complex event processing system that performs queries very quickly on incoming data. Now Boston-based EnerNoc is seeking to expand into a new line of business—energy-management services for commercial building owners—and analytics is central to it, says Dixon. Its software will collect data from 6,000 points in a building every five minutes to generate monthly recommendations on how companies can run their buildings more efficiently. By comparing one building’s energy profile to others, it can be more accurate in predicting potential energy savings and recommend times to reduce electricity use.
The ultimate goal is to use data from meters and building management systems to send signals to building control systems, Dixon says. “Right now the technology has evolved to the point where the leading edge is that you can automatically find inefficiencies but only some of them can you automatically correct,” he says.
The Instrumented Data Center
Utility meters and buildings are one common example of machine data streams that companies are trying to capitalize on. But there are a growing number of examples of what is called the Industrial Internet or the Internet of things.
Phoenix-based IO makes a modular data center, a self-contained data center in a shipping container-size box. To differentiate its product from other modular data centers, it has developed a software management application that allows for fine-grained control.
Every piece of equipment in the modular data center is instrumented, creating a flow of information that populates a database that’s already 12 billion rows long, according to Kevin Malik, the CIO of IO Labs, the company’s research arm. That includes sensors to track the temperature of the server processors, the speed of air flow in the air handler, and the amount of power being provided. Each modular data center has 83,000 sensor points where data is collected every five seconds.
This real-time telemetry allows data center operators to spot potential problems, such as failed hardware, or make adjustments to tune performance, such as consolidating workloads to save energy. The technical challenge for IO in writing the software was not so much collecting the information, but turning it into something meaningful, says Malik. “Does the data have relevance or not? Unfortunately, it means collecting as much data as you can and sorting through it,” he says.
IO invested heavily in the user interface design and visualization tools to make the data understandable to technology pros. Its control dashboard application shows a globe where people can view different data centers and then drill down to see performance information at a specific site, or specific component within a data center. To help make sense of the flow of information, the software creates heat maps and alerts to indicate where there are potential problems and anomalies.
“The next thing is predictive analysis. Because we have this data and we’ve written these algorithms, we can send you an alert on the dashboard or your phone that there’s a 95 percent change that this air handler will fail in the next three weeks,” Malik says. “And if I can turn things on and off (from the dashboard), I have tremendous power in my hands.”
Vertica, which makes a columnar database optimized for real-time analytics, is finding that a number of industrial customers are starting to make use of sensor data, according to Jeff Healey, the director of product marketing, who hosted a webcast on the Internet of things on Feb. 14. Power plant operators and oil drillers, for example, can get vital statistics on their physical infrastructure, such as pipelines or cooling towers. A plane can generate 2.5 terabytes of data in one year, data that can be analyzed to help improve aircraft safety.
One of the main reasons companies are using telemetry and sensors is to reduce down-time of expensive equipment, such as medical devices in hospitals or servers in a data center, Healy says. Software can also differentiate commodity hardware by creating services around it. For example, a printer can send an alarm based on machine data to indicate that ink is low which would then trigger an email offer to consumers at a discount.
As more data from sensors and other control equipment becomes available, expect more specialized technologies to emerge for collecting data from multiple formats and analyzing it. Vertica’s Healey predicts more companies will find more use cases, or business reasons for analyzing machine data, over time. “Sensor data is generating enormous amounts of data. It’s going to make big data extremely big, particularly as people understand by analyzing this data you fuel a lot of different use cases,” he says.
Martin LaMonica is a technology journalist in the Boston area. Follow him on Twitter @mlamonica.