Glossary · PLC and control programming
TwinCAT Machine Learning
Also known as: TwinCAT ML, Beckhoff machine learning inference
German: TwinCAT Machine Learning
In industrial automation, TwinCAT Machine Learning is Beckhoff's TwinCAT functionality for executing trained machine learning models, for example models exchanged in the ONNX format, within the TwinCAT runtime so that inference can run in real time alongside the control program.
- PLC programming
- Vendor product
- AI
In one sentence
TwinCAT Machine Learning runs trained ML models, for example in ONNX format, inside the TwinCAT runtime alongside the control program.
Example
A model trained on motor current data runs in the TwinCAT runtime and classifies each machine cycle as normal or suspicious.
How it applies
- Engineering: Models are trained outside the controller with common ML frameworks, exported and loaded into the runtime, where the PLC program calls them. Training and Training vs. inference are therefore separate steps with separate tools.
- Validation: A model's behavior depends on its training data. Validation must cover the operating range of the machine, and changes to the model are software changes that need testing and version control, see Model governance.
- Regulation: If an ML function takes over a safety function or is otherwise covered by product regulation, the requirements of the Machinery Regulation and, where applicable, the AI Act apply. See AI-enabled machine function.
- Documentation: Record the model version, training data description, input signals, output meaning and limits. Users need to know what the model decides and what happens when it is uncertain.
Keep in mind: product names and features change. Check supported model formats and functions against the vendor's current documentation.