Machina Aeternum : Predicting the Unpredictable

Airvac, Inc. - the vacuum sewer company headquartered in Rochester, Indiana - is the latest production customer for Machina Aeternum. They design and support vacuum collection systems worldwide, and they offer a white-glove monitoring service that watches field devices for the utilities.

Over 53,000 sensors and more than 1,000,000,000 data points

Until this deployment, that service meant reviewing every monitored device by hand, every day. Machina Aeternum now reads the sensor streams Airvac already collects, flags events, and scores device health so reviewers start with the assets that matter.

In addition to monitoring, Machina Aeternum predicts future events and device health.

Machina Aeternum is Peakey Enterprise's flagship predictive maintenance platform. Built on our technology first introduced in our 2021 whitepaper "Predicting the Unpredictable," it combines multivariate time-series forecasting with intelligent anomaly detection to identify degradation, failures, and "false anomalies" - events that look random to humans but are actually predictable when all sensor relationships are modeled together. Unlike traditional threshold alerts or basic anomaly detectors, Machina Aeternum:

• Pulls data securely from platform, database, IoT, or custom implementation. (streaming or batch)
• Forecasts future sensor values across multiple streams
• Runs anomaly detection on both current and predicted future states
• Scores device health from 0-100, with event penalties that roll off over time
• Triggers automated alerts (email, SMS, automatic shutoff, or any custom implementation) when issues are detected
• Supports a hybrid rules engine: machine learning plus your hard-coded logic for events that must never slip.
• Escalates low-confidence events to your experts for labeling (human-in-the-loop Review).

All processing can run on your infrastructure when data sovereignty or offline operation is required.

• Real-time / Batch Anomaly Detection: Flags deviations as they occur.
• Multivariate Forecasting: Predicts how multiple sensors will behave together over configurable time horizons.
• Predicted events and predicted health: Visible in the product and in reports. Customers put them into dispatch when they choose.
• Human-in-the-loop: Low-confidence events go to your experts, with labeled examples for comparison.
• Hybrid Rules Engine: Combine ML with your custom hard-coded logic for critical events.
• Extensible & Integratable: Easy connection to databases, notification systems, and ticketing tools for model monitoring.

• OEM and service teams that watch someone else's installed base
• Owner-operators watching their own production lines, fleets, or plants
• Operations that do not yet have a usable feed - Peakey can help specify sensors and devices, then run Machina Aeternum on that stream.

• Up to 50% reduction in unplanned downtime
• 25-40% lower maintenance costs
• Improved safety, reduced scrap/waste, and higher asset reliability

Real-world example outcomes include preventing machine failures, maintaining quality thresholds, optimizing throughput, and avoiding service outages on everything from industrial valves to fleet equipment.

Machina Aeternum builds directly on Peakey's (Predicting the Unpredictable: Implementation of an Advanced Machine Learning Method to Predict Degradation and Failure using Multivariate Time-Series Forecasting @ 2021).
Our approach solves two major challenges in industrial AI:

• Self-supervised learning: No massive manual labeling required upfront.
• False Anomaly Prediction: Identifies events that appear anomalous but are predictable when all dependent variables are modeled together.

In independent testing on public industrial datasets (e.g., SKAB valve data), our models successfully predicted sharp anomalies - such as a dramatic drop in flow rate RMS - nearly 10 minutes in advance using only seconds of prior data.

• Runs fully offline on your servers when you need it - or hosted by Peakey
• Field devices stay in many deployments. A source-specific import adapter is the usual engineering work, quoted in the pilot. Once that adapter exists for a source class, more devices of that type are configuration.
• Hybrid ML + rules approach ensures nothing critical slips through
• Predicted events are in the base product. You decide when they hit the dispatch board
• Built and supported onshore in Warsaw, Indiana USA

A pilot is a parallel run, not a rip-and-replace. Typical start: one device type, 30-50 assets, a written definition of success, and a quoted import adapter against your data path. Contact us today to discuss a pilot or request the full whitepaper.