Advanced Prognostics AI-MSET™
Based on the industry’s leading Prognostic Algorithm, MSET (Multivariate State Estimation Technique) predicts in real time what each process metric should be by learning correlations among process variables across subsystems, large-scale assets, or entire plants with thousands to millions of sensors.
This approach combines advanced pattern recognition with SPRT (Sequential Probability Ratio Test), which allows independently configurable false- and missed-alarm probabilities. Together, they provide highly sensitive, proactive detection of subtle anomalies in complex, noisy process variables while maintaining extremely low false alarm rates.
Background and Applications
MSET was originally developed by Argonne National Laboratory (ANL) for high-sensitivity proactive fault monitoring in nuclear power systems and has since been adopted globally for streaming prognostics in nearly all nuclear plants worldwide. Over time, its use has expanded into a wide range of mission-critical and safety-critical applications — from NASA’s space programs and military turbine diagnostics to oil and gas asset prognostics, industrial and manufacturing systems, commercial aviation, and theme park structural reliability systems.
For the last two decades, significant advancements have been made — pioneering over four dozen patented algorithms that extend and enhance the original MSET and SPRT framework into what is now known as AI-MSET™, a real-time prognostics system designed for early anomaly detection in business-critical computing, industrial, and infrastructure environments.
Key Capabilities
AI-MSET™ is designed to identify the smallest developing faults as early as possible, enabling operations teams to respond before issues escalate. It detects early signs of:
- Sensor degradation and sensor decalibration mechanisms
- Onset of bearing out-of-roundness, chipped gear tooth, radial imbalance in centrifugal pumps
- Shaft binding, lubrication, dry out in rotating machinery
- Winding irregularities in oil-filled transformers
- Onset of corrosion and fouling in flow circuits
- Structural and process anomalies in mechanical and electromechanical asset
AI-MSET™ integrates four pattern recognition and optimization modules that outperform conventional machine learning techniques such as neural networks, kernel regression, and support vector machines — offering higher prognostic accuracy, lower false and missed alarm probabilities, and far lower computational cost.
The system includes both a training phase(characterizing normal operation using historical error-free data) and a monitoring phase (comparing real-time data against the trained model). Its estimates maintain uncertainty bounds of just 1–2% of the raw sensor standard deviation, delivering exceptional sensitivity and reliability for anomaly detection.
Remaining Useful Life (RUL)
The Remaining Useful Life (RUL)component provides highly accurate asset life estimations with a quantified confidence factor. This capability enables operators to transition from time-based or condition-based maintenance to fully predictive asset management strategies.
Utility Sector Use Cases
AI-MSET™ supports a wide range of utility and energy applications:
- Fraud Detection (Meter Tampering, Meter Swapping, Theft of Electricity)
- Predictive and preventive maintenance for all types of grid and plant assets
- Energy efficiency and demand response
- Strategic asset management and capital planning
- Cable fault detection (overhead and underground)
- Transformer predictive and prescriptive maintenance
- Substation and SCADA proactive fault monitoring
- Highly Accurate Load Shape Forecasting
- Optimal Resource Allocation for Minimizing Power Outages from Storm Events
- Prognostic Cyber Security of SCADA Systems & Networks
Advantages Over Threshold-Based Monitoring
Threshold monitoring is not predictive…it is reactive, because by the time a threshold is crossed, the degradation mechanism is already well underway. Tighten the Thresholds, and the number of false alarms increase, which wastes time and money shutting down revenue-generating assets unnecessarily. Loosen Thresholds to avoid false alarms, and missed alarm probabilities increase, meaning the degradation can be severely underway before any alert is generated (or the asset can be failed). AI-MSET™ proactively detects the onset of incipient developing faults so that small anomalies can be proactively repaired before turning into an expensive asset failure.
AI-MSET™ attains extremely high sensitivity for detecting subtle disturbances in noisy process variables, but with extremely low false positives and false negatives.
Maintenance and Capital Optimization
AI-MSET™ automatically correlates and synchronizes telemetry data from diverse sources — such as DGA, SCADA, smart meters, field inspections, and weather data — across vastly different sampling frequencies. This produces the most accurate and reliable RUL estimates in the industry, supporting efficient maintenance planning, optimized capital allocation, and reduced operation and maintenance (O&M) costs.
Performance Benchmarks
Every AI-MSET™ implementation can be measured against four key quantitative criteria for time-series prognostics:
- Earliest anomaly detection —weeks or months ahead of threshold-based systems
- Lowest false alarm probability (Type I error)
- Lowest missed alarm probability (Type II error)
- Highest Computational Efficiency —essential for dense-sensor and/or high-frequency data-acquisition environments. AI-MSET™ attains 3 orders of magnitude lower compute costs than neural-network based monitoring for the same numbers of sensors and same sampling rates.