AI-MEST Technology

---

The industry’s leading prognostic algorithm, AI-MSET, predicts in real time what each process metric should be by learning correlations among process variables. It can be applied to subsystems with dozens of sensors, large-scale assets with thousands, or entire plants involving millions.

Prognostic innovations combine advanced pattern recognition with the SPRT (Sequential Probability Ratio Test), which allows independently configurable false-alarm and missed-alarm probabilities. This enables high sensitivity for proactively detecting subtle anomalies -even in noisy process variables - while maintaining ultra-low false alarm rates.

AI-MSET is engineered to detect the earliest signs of developing faults, giving operations personnel advance warning of sensor degradation and system anomalies such as shaft binding in rotating machinery, corrosion, or flow circuit fouling. It integrates four advanced pattern recognition and optimization modules that exceed neural networks in sensitivity, reliability, and computational efficiency.

Most IoT systems rely on static thresholds, which present inherent challenges. Once an alert or “trip” occurs, the notification often comes too late - damage has already begun or failure is imminent, leaving only hours of reaction time. Tightening thresholds increases sensitivity, but also leads to a surge of false alarms. This overwhelms operators, leading to alarm fatigue or wasted resources when maintenance crews are dispatched only to find “NTF” (No Trouble Found).

Unlike threshold-based systems, AI-MSETdynamically creates a real-time band around each sensor value and automatically correlates it with other sensor data (including field inspection inputs). Leveraging the SPRT (Sequential Probability Ratio Test), it minimizes false alarms while achieving extreme sensitivity to distinguish between simple sensor disturbances and genuine asset failures. This provides the earliest possible warnings for IoT applications.

AI-MSET offers substantial advantages over conventional machine learning methods—including neural networks, auto-associative kernel regression, and support vector machines—by delivering higher prognostic accuracy, lower false- and missed-alarm probabilities (FAPs and MAPs), and significantly reduced computational costs. This efficiency is critical for real-time prognostics in dense-sensor environments.

The AI-MSET framework consists of two stages:

• Training Phase: Uses historical, error-free operational data across the full range of system operating conditions. An optimal subset of data (memory vectors) is selected to best represent normal asset behaviour, forming a reference model.
• Monitoring Phase: New incoming sensor observations are compared against the trained AI-MSET model to estimate expected signal values. These estimates are highly precise, with uncertainty bounds typically only 1–2% of the standard deviation of the raw sensor inputs.  

Through this approach, AI-MSET achieves unparalleled sensitivity for detecting subtle anomalies in noisy process variables, while maintaining extremely low false positives and false negatives.

*IEEE published and peer reviewed. Provides the earliest possible detection of any ML technique of anomalist behaviour without false or missed alarms

^ Has the lowest possible occurrence of false and missed alarm probabilities of any ML technique in the industry today which has also been validated by IEEE, published and peer reviewed. 1 x 10^5 over 10,000 hours or about 1.2 years.

& Requires 2-3 Orders of Magnitude Less than Neural Networks (LLM, LSTM, Convolutional, etc.). Can process Thousands of Signals with Observation Rate at Hertz on a Raspberry Pi W Zero 2. Example: 300,000+ signals, 1 year of data with 1 hertz observations on a laptop in less than 20 minutes. 

BUSINESS IMPACT

• Prevents unplanned downtimeand catastrophic equipment failures
• Lowers operational costsby reducing unnecessary maintenance and false shutdowns
• Accelerates time-to-valuewith minimal integration friction.
• Enhances reliability and safety across high-stakes industrial and IT environments