Predictive maintenance is critical for industrial organizations to maintain safe and productive operations. Traditional methods of machine condition monitoring (MCM) offer reliable coverage but are configured from human experience and rules of thumb, making them susceptible to unique, never-before-seen changes in asset behavior. If not properly detected, these events cause suboptimal operations or even catastrophic failures that ruin operations and maintenance budgets. By leveraging asset data, maintenance teams improve their operational productivity by supplementing their existing maintenance strategy with AI-based predictive analytics. SparkPredict® leverages live-streaming data from the OSIsoft PI System and high-performance machine learning models to help maintenance teams identify, assess, and troubleshoot never-before-seen behaviors while incorporating subject matter expertise into a monitoring system that improves over time.
AI-based analytics present a significant business opportunity to reimagine industrial maintenance with practices that reduce unexpected downtime and optimize maintenance workflows. Customers can directly translate the benefits of these analytics to their bottom line by reducing risks of catastrophic failure and maximizing effectiveness of their teams.
We’ve worked with leading industrial companies to:
- Prevent catastrophic failure by identifying never-before-seen behavior on critical gas turbines
- Improve forewarning of shipboard propulsion motor failure by 20X
- Reduce alarm false positives by 80% in hydroelectric power plant
- Generate a new revenue stream for a gas turbine OEM
SparkPredict pulls data directly from the OSIsoft PI System using a variety of techniques including Asset Framework and PI AF SDK. The archived sensor data is analyzed with predictive machine learning models to generate accurate alarms and insights. An SME can take advantage of the diagnostic capabilities within the SparkPredict UI to investigate alarms with a high level of precision using capabilities such as advanced tag analysis, time scaling, and cluster comparison. When new patterns are identified, the SME can label them from within the UI and then retrain the model to recognize the pattern automatically in the future. With this approach, the machine learning model gets more accurate over time and results in improved asset performance.