Download Video: MP4, WebM, Ogg

Machine learning for all areas of automation

Beckhoff now offers a machine learning (ML) solution that is seamlessly integrated into TwinCAT 3. Building on established standards, it brings to ML applications the advantages of system openness familiar from PC-based control. In addition, the TwinCAT solution supports the execution of the machine learning models in real time, allowing it to handle demanding tasks like complex motion control. Its capabilities provide machine builders with an optimum foundation for enhancing machine performance.

The fundamental idea with machine learning is to no longer follow the classic engineering route of designing solutions for specific tasks and then turning these solutions into algorithms, but to enable the desired algorithms to be learned from model process data instead. Beckhoff offers a closed workflow for the entire circulation of data and data-based algorithms and deployment.

The workflow from data collection, through training, to integration of the trained model into the TwinCAT 3 Runtime (XAR)

The workflow from data collection, through training, to integration of the trained model into the TwinCAT 3 Runtime (XAR)

Real-time-capable ML model inference capabilities as a standard module in TwinCAT 3

Real-time-capable ML model inference capabilities as a standard module in TwinCAT 3: No extra hardware is required. The functionality is implemented entirely in software on the same standard platform as the other control applications.

Each application places different demands on the method of archiving machine data: SQL or noSQL, file-based, cloud-based data lake, etc. For all these scenarios there are a large number of established TwinCAT products available, such as the TF6420 TC3 Database Server, the TF3300 TC3 Scope Server or the TF6720 TC3 IoT Data Agent.

Training is performed in established frameworks such as PyTorch, TensorFlow, SciKit-Learn, MATLAB®, etc. This ensures maximum flexibility. No limits are set in the case of an interdisciplinary project between automation engineers and data scientists – neither internally in the company nor beyond the bounds of the company. The learned model can simply be exported in a standardized format (ONNX) and handed over to the TwinCAT programmer.

Deployment takes place via TwinCAT Engineering directly into the TwinCAT XAR, so that the learned model (inference) is executed directly in hard real-time on the machine controller and is thus synchronous with all other controller objects.

In automation technology, this opens up new possibilities as well as optimization potential in such areas as predictive maintenance and process control, anomaly detection, collaborative robotics, automated quality control, and machine optimization.

Products

TF3800 | TwinCAT Machine Learning Inference Engine

TF3800 | TwinCAT Machine Learning Inference Engine: execution module of trained classical machine learning algorithms

TF3810 | TwinCAT Neural Network Inference Engine

TF3810 | TwinCAT Neural Network Inference Engine: execution module of trained neural networks

Further information