Artificial Intelligence
Anomaly detection, signal classification, predictive maintenance and flight data analysis on sensor data, including lightweight models that run on embedded hardware.
What we do
We develop artificial intelligence models that turn sensor data into actionable information: anomaly detection, signal classification, predictive maintenance and flight data analysis fall within this scope.
Models are often designed to run on embedded hardware with a limited power and memory budget; model choice and optimisation are therefore handled together with the constraints of the target hardware.
The work is not limited to a one-off model handover: what data the model was validated against, and with which acceptance criteria, is also part of the delivery.
Capabilities
- Anomaly detection on sensor data
- Signal classification
- Predictive maintenance models
- Flight data analysis
- Lightweight model design for embedded hardware
- Model validation and acceptance criteria definition
How we work
Data assessment
Existing sensor data and the target output are reviewed together to determine the right approach.
Model design
A model architecture suited to the task is selected or designed.
Embedded optimisation
The model is reduced to fit the target hardware's power and memory budget.
Validation
The model is evaluated against defined test data and acceptance criteria.
Delivery
The validated model is integrated into the relevant hardware or software system and delivered.