A year ago, I wrote about a fairly new company, SparkCognition, that had figured out a way to harness and make great use of the mountain of data spewing forth from our ever-increasingly instrumented society. Since then, the company has made swift progress.
Amir Husain, SparkCognition’s founder and CEO, has advanced the state of cognitive computing so fast it’s hard to keep up. For his efforts, he has been recognized in a number of forums. On Nov. 5, he was named one of six “Tech Titans for 2015” by the Austin Business Journal. The company was also featured in the November issue of Inc. A few days after the Austin Business Journal piece, Husain headed to Helsinki to accept an award at Nokia’s IoT Open Innovation Challenge.
SparkCognition focuses on two areas, cyber security and the industrial Internet. The degree of traction it has gained in the latter is remarkable.
The industrial Internet refers to the all the complex machinery that runs our infrastructure, equipment that, in today’s world, is increasingly interconnected. These machines — gear like oil and gas pipelines, wind turbines, power generators, water distribution systems, process manufacturing facilities — are usually expensive, often huge, and in many cases surprisingly sensitive to things going even just slightly wrong. Witness the humble ball bearing. A crack in this tiny part can take out a $10 million turbine.
Many equipment makers have been putting sensors on their machines for some time now with the idea in mind that data from these sensors is valuable. The problem has been that, in many cases, the sensors haven’t been hooked up to anything, and if they have, it’s typically something simple like a “reader.” So, if a turbine has a vibration or temperature sensor, there’s some way to look at that information, but no way to see that information in a larger context.
What SparkCognition has done is create a way to gather all this information, put it into a database, comb it for meaning, and weigh its component parts. Is it one temperature gauge rising, all of them, or only some of them? Is the information from this particular gauge meaningful, or is some other gauge surfacing more useful information? How is the information from one gauge related to others? Has this happened before? What else is happening in the system? Anything unusual?
Using reasoning like this, SparkCognition software generates models that might produce the observed effect. It then tests these models to see if in fact they do produce such behavior, creates confidence rankings for the various promising models, and offers the one with the highest ranking as the likely cause-and-effect map. This model can then be used to predict the effect based on real-time inputs. All while not “knowing” a thing about the actual system involved. In other words, the software doesn’t have to embody the expertise of a power-generation engineer; it just looks at the data itself.
The opportunity is enormous, particularly in the energy area, which comprises 5–6% of U.S. gross domestic product (GDP) and above 10% in much of the developing world. With world GDP valued at more than $100 trillion, according to a CIA World Factbook estimate at purchasing power parity, the energy sector represents, conservatively, a $5 trillion annual spend. In an industry this big, even tiny increases in efficiency are a huge win.
And there are many tiny increases to be made in the efficiency, safety, and security of power generation and distribution. Technology offers the scalability required for large, spread-out operations.
For example, an oil company in Mexico has set drones and other sensors to monitor for oil pilferage along its pipelines. But monitoring the output of all these field devices is impossible manually. Smart systems need to pull it all together. And there are billions of dollars of assets like these in Central Asia, Russia, China, and elsewhere. In the near future, they will all have sensors, and they all need to be secured.
But security is only one aspect of this massive industry. Maintenance, although more pedestrian, may actually represent a larger opportunity for savings. And direct efficiency could be the largest opportunity of all. Energy generation must become more efficient, reducing the climate impact of fossil fuel. Utilities need to be more efficient all around the world. This is one of the biggest economic themes in both developed and developing countries.
Husain calls the job of hooking all these sensors up to an aware system an “8 to 10 year challenge” and labels the opportunity to monitor and make sense of all the data these systems will generate a “digital oil field” for companies, like his, able to drill for it.
SparkCognition has already changed the status quo in some industries. One of its early successes was with Flowserve, which makes industrial pumps and related equipment. Keeping up with the times, Flowserve now offers an Amazon-like Web-Service for improving asset performance for process industry companies in domains like energy consumption, equipment reliability, inventory efficiency, and maintenance. The service is designed to monitor remote sensors for identification, diagnosis, and recommendations to improve equipment reliability. With SparkCognition integrated into its own software, Flowserve can create a full picture of maintenance-related events, helping to transform the customer from a reactive to a predictive footing. This type of efficiency is not just some abstract notion. The SparkCognition implementation has helped Flowserve increase its advanced notice on pump failure from hours to days. That’s five days’ notice on an impending failure of an expensive and potentially remote asset.
One of the largest wind farms in North America uses it to predict wind turbine failures and for fleet optimization. Fleet optimization is an example of a direct efficiency use case. Turbines are not isolated, but stand in fields with many others. In some cases, a turbine upwind of another can increase the efficiency of the downwind turbine if it operates outside its own highest efficiency speed. Only a system aware of the sensor outputs of both can make this adjustment.
Another customer, a major U.S. utility, is using SparkCognition technology to look for startup anomalies in turbines that cost $25 million each. Other customers include one of the foremost oil field service companies, an industrial controls company, several software companies, a major computer hardware company, sensor companies, and the U.S. government. And no litany would be complete without mentioning that one of the firm’s early partners, IBM’s Watson Group, has incorporated the technology into its analytics toolset. SparkCognition allows all these companies to offer “everything as a service,” as Husain puts it.
In a year of selling, the nearly-startup has closed 30 enterprise customers, all the more remarkable because large companies often have longer sales cycles. The firm continues to add high-end talent to its roster, including people in operations research, statistics, and physics.
There is no contesting that SparkCognition is getting rapid traction as it focuses down in the industrial Internet and the energy sector in particular. The company continues to raise capital in the private markets.