📩 Aperture@thegaogroup.com
Industrial Digital Twins and Simulation
Physical AI and AIoT Research Team of
Aperture Venture Studio, GAO RFID Inc. and GAO Tek Inc.
This program provides experimental and decision-support infrastructure across the other programs.
Operational Digital Twins for State Estimation and Action Evaluation
Build synchronized representations of selected equipment or processes that combine physical models with live sensor data. Use these twins to estimate hidden states, diagnose problems and evaluate candidate actions.
The research contribution should concern model accuracy, synchronization or decision usefulness, rather than visual appearance alone.
Research questions
How can a twin remain calibrated as equipment changes? How accurately must it predict an outcome to support a particular decision?
Initial demonstration
A pump, motor or thermal-system twin that compares predicted responses with physical measurements.
Evaluation
Prediction error, synchronization delay, uncertainty calibration and improvement in decisions supported by the twin.
Simulation-to-Reality Transfer and Synthetic Industrial Data
Investigate how simulation and synthetic data can reduce the amount of physical experimentation needed to train or evaluate models.
Potential applications include inspection under unusual conditions, robot perception and control-system testing. Simulated results must be checked against independent physical observations.
Research questions
Which simulated variations improve real-world performance? How can the system detect when simulation assumptions do not represent the deployment environment?
Initial demonstration
Train an inspection or control model using simulated data, then evaluate it on a physical testbed.
Evaluation
Real-world performance, required physical training data, transfer degradation and coverage of difficult operating conditions.
Industrial Sensor Reliability and Data-Quality Intelligence
Develop methods to determine whether incoming measurements are trustworthy before they influence an AI decision. Address sensor drift, calibration errors, communication gaps, incorrect units, timestamp misalignment and device substitution.
This deserves its own topic because a system must distinguish a failing machine from a failing sensor. It also provides a reusable capability across Aperture’s portfolio.
Research questions
Can redundant measurements and physical relationships reveal faulty sensors? How should unreliable data affect model confidence and permission to act?
Initial demonstration
Introduce sensor faults and missing readings into an instrumented test rig.
Evaluation
Sensor-fault detection, false alarms, fault-isolation accuracy and downstream decision quality.