5 Years of SparkCognition, 5 Problems We’ve Solved

This week, SparkCognition celebrated its fifth year as a company. In that short time, we have emerged as a leader in the field of artificial intelligence, winning awards for our technology and gaining national attention for our growth. But as a team of problem-solvers, what’s most meaningful for us is bringing innovative research out of the lab and translating it to real-world use for companies that advance the most important interests of our society.

Thus, for our fifth birthday, we wanted to share five examples of SparkCognition technology at work in these critical industries. Here’s a look at the solutions we’re providing, from our first use case to our latest innovation.

Click on each title to read the full case study.

A major utility invested in SparkCognition, calculating that it would pay off it SparkPredict caught at least one significant, undetectable issue over three years. In fact, SparkPredict helped avert a catastrophic issue in the first 4 months.

Problem: A major utility wanted to protect a $200-$400M turbine investment from critical unknown defects

Solution: The company leveraged SparkPredict to analyze live-streaming turbine data and identify new failure indicators.

Result: With SparkPredict, the utility avoided a unique, catastrophic event that went undetected by the monitoring systems in place, with one month advance notice.

Hydro turbine units are highly reliable, meaning that few examples of unplanned downtime exist. However, due to the high cost of failures, reliable predictive analytics represented a significant opportunity for the industry.

Problem: A hydro utility had an unplanned outage with an estimated $1.5M impact.

Solution: The company evaluated how machine learning could predict uncommon failures with sparse data.

Result: SparkPredict identified the large scale outage with one month advance warning.

In a business where unexpected failures and unscheduled downtime are both costly and dangerous, the ability to predict failures in critical assets before they occur is paramount. An E&P (exploration and production) company needed a scalable solution to rethink maintenance.

Problem: An E&P operator needed to monitor the state of all the wells in a field and prioritize maintenance based on well production.

Solution: The operator utilized Darwin™, SparkCognition’s automated development platform, to build optimized models applicable to all wells in the field that identified maintenance needs and potential return of individual wells after service.

Result: Darwin-generated models helped process engineers predict workover, rod change, and cleaning operations needs in 12 out of 17 wells across the field with 70-80% accuracy. This provided a scalable solution in a short timeframe with millions of dollars in estimated cost savings.

In the maritime industry, going off-hire due to an unplanned out-of-service period can result in an opportunity cost of up to several hundred thousand dollars per day, not to mention the negative publicity of a ship stranded at sea.

Problem: An industry-leading ship operator had no predictive abilities for failures, and after a significant incident, decided to evaluate solutions with a goal of 2 weeks’ advance notice.

Solution: SparkPredict used sensor data from engine room assets to identify the 8 main signals indicating a potential failure.

Result: SparkPredict predicted failures in propulsion motors up to 10 months in advance, and failures in alternators up to six weeks in advance. SparkPredict was even able to identify a imminent failure on ancillary equipment, even without data from that part.

The icing on the SparkCognition cake! Darwin is a prime example of SparkCognition translating cutting-edge research into the real world. In fact, our data scientists will be presenting a paper on automated model building at the prestigious GECCO conference next month.

Problem: With an urgent need to make data-driven decisions and current machine learning builds taking weeks to months, companies need models faster.

Solution: Darwin automates the complete data science process, for both supervised and unsupervised learning problems, to empower existing talent and activate the value of data.

Result: Darwin’s models were consistently more accurate than other methods of model-building, some achieving a perfect fit with a runtime of less than 20 minutes (when other methods weren’t even close).

With every meeting, interview, article, blog post, and of course, case study, we strive to bring artificial intelligence out of hypotheticals and into the real world. Thank you for following along on SparkCognition’s journey. Here’s to the next five years!