Data & Integration
Sensor history as training data: first deployments
What we learned when we plugged a world model into real SCADA historians and saw the data that actually comes out.
Research and engineering notes
From the AMI Labs team on equipment world models, sensor data, process control, and on-premise industrial AI.
World Models
Why process engineers need a machine twin
Every setpoint adjustment is a live experiment. We argue that a learned model of your specific machine changes the equation.
Process Engineering
The hidden cost of undocumented process changes
When an operator changes a set point and does not log it, the historian shows an anomaly with no explanation. Here is how we detect and correct for undocumented changes.
World Models
World models vs. regression for equipment behavior
A direct comparison of classical regression approaches versus causal world models when predicting how industrial equipment behaves across multiple operating regimes.
World Models
Vibration signature fingerprinting for bearing health
How the world model learns the baseline vibration fingerprint of a specific bearing, and what deviations from that fingerprint indicate about bearing health.
Process Engineering
Control loop tuning, simulation first
PID tuning traditionally requires live trials. We walk through how a machine world model changes that.
Data & Integration
Data collection checklist for an equipment world model
Not all sensor history is equal. Here is what to check before you feed data to a world model trainer.
Data & Integration
From SCADA historian to learning system
What needs to happen between extracting raw historian data and having a usable world model. A practitioner walkthrough.
Technical
Predicting setpoint change effects: a technical walkthrough
The causal inference engine at the core of AMI Labs: how it represents control variables, state, and response trajectories for forward prediction.
Technical
Benchmarking equipment world model accuracy
How we measure whether a learned model of a machine is accurate enough to trust for live prediction.
Safety
Thermal runaway prediction in batch reactors
A world model trained on temperature and feed-rate history can identify the precursors to thermal runaway before a human operator can detect them.
Data & Integration
OPC-UA as the data backbone for real-time inference
Why OPC-UA is the right standard for feeding live sensor data to an equipment world model, and how we integrate it into on-premise inference.
Technical
Edge deployment of equipment world models and latency tradeoffs
On-premise and edge deployment changes the prediction latency profile significantly. Here is what matters and what does not.