A cognitive approach to industrial safety and efficiency

SparkPredict® Raises Machine Prognostics to a Cognitive Level

SparkPredict® enables truly predictive capabilities that deliver billions of dollars in cost savings and operational efficiency improvements to machine operators. SparkPredict® learns from sensor data, identifies impending failures long before they occur, and flags sub-optimal operations before they can cause any harm.
Existing machine condition monitoring systems are good at looking at individual variables and monitoring when each variable breaches its assigned thresholds. Some rely on legacy approaches such as mean time between failures (MTBF) to schedule interventions such as service and replacements. These methods simply aren’t very good at predicting what will happen next in a manner that’s timely enough to make a real difference. SparkPredict® is!


Improved Precision

Sophisticated algorithms exceed accuracy of traditional methods


Faster Results

Automated model building shrinks days or week-long model building processes to minutes



Minimum human engagement enables large-scale deployments



Automated model building also enables the algorithm to adapt to specific maintenance schedules, load patterns, and environmental conditions



Self-learning and adaptive algorithms ensure that results continue to improve over time.

How SparkPredict® works

SparkPredict® applies sophisticated algorithms, including our own patent-pending cognitive algorithms, to huge bodies of sensor data and big analog data generated by pumps, rotors, motors and a variety of other machines. SparkPredict®:
1. Collects large amounts of sensor data
2. Acquires and transforms the data
3. Applies SparkCognition’s proprietary classification and prediction algorithms
4. Generates actionable insights and supports root cause analysis

Costly Failures

In the United States, one of the leading causes of machinery malfunction is bearing failures in rotating assets such as generators, actuators, turbines, and motors. According to an MIT study, more than $240 billion is lost in downtime and repair costs annually.

Proprietary Algorithms

These are just a few of the sophisticated algorithms at work in SparkPredict®. In combination, the algorithms power a platform that allows our customers to quickly deploy against all of their assets, automatically find anomalous behavior, and build asset specific models that predict or classify on an ongoing basis without any supervision.
Artemis™ autonomously builds, tests, and identifies meaningful relationships from thousands of combinations. The technology behind Artemis is especially suited to finding anomalies–not just single-variate anomalies, but anomalies based on combinations of variables as well.
Pythia™ is an automated model building algorithm. It develops functions that “learn” the physics of underlying systems or assets from data feeds. Pythia uses a variety of machine learning techniques to initially distinguish more obvious noise from “signal.” After this first order filtering, the algorithm combines deep learning and genetic algorithmic methods to develop models that predict when the machine is likely to fail.

See how SparkPredict® can protect your infrastructure