From Machine Signals to Maintenance Decisions: Enabling Predictive Sustainment at Scale
Customer Challenge
The Air Force required a scalable capability to deliver predictive maintenance support for Industrial Internet of Things (IIoT) systems and machines.
Innovative Solution
Illumination Works trained AI/ML models using IIoT sensor data (930 sensors) to detect early signs of machine failures. These models identified machine health fault signatures in real-time leveraging two sources of IIoT data: scalar magnitude (5-second level) and spectral frequency (millisecond level). Early machine fault detection using AI/ML allows the Air Force to make informed, real-time decisions regarding preventative, condition-based, and predictive maintenance strategies, which ultimately improves the bottom line of large-scale infrastructure required to efficiently operate hundreds of high-cost machines.
Benefits/Outcomes
- Applies highly performant AI/ML models to enable predicting machine failures more accurately and faster than traditional predictive maintenance workflows
- Outperforms human-led decision cycles using large quantities of unstructured, granular IIoT spectral frequency data collected at the millisecond level (2.1 billion records)
- Improves data-informed, real-time decision making via visualization of a broad range of complex machine fault signatures (e.g., imbalance, misalignment, bearing fault)
- Demonstrates efficient management of highly granular real-time sensor data
Business Value
- Improves operational readiness, machine availability, downtime, and personnel productivity
- Informs the Capital Investment Program (CIP) that allocates money to key areas, like replacing equipment
- Drives substantial cost savings for industries with expensive infrastructure and operations
- Enables faster and more informed decision making leveraging real-time compute processing with AI/ML
Toolbox
- Modern cloud platforms: Microsoft Azure and AWS
- Cloud-based tools: Databricks, PySpark, MLflow
- Medallion Architecture, Data Lakes, MongoDB
- Time-Series Forecasting, IIoT Big Data Processing, MLOps
- Data Visualization using MLflow, Databricks, and Tableau
Domain Expertise
- Industrial machine data
- Vibration sensor data
- Predictive maintenance
