The TF3820 Twincat 3 Machine Learning Server by Beckhoff is an inference engine for trained ML and DL models whose programming takes place in the PLC. From here, models can be loaded, the executing hardware configured and the inference executed. The execution of the model (clustering, image classification, object detection…) takes place in an independent process of the operating system. The training is done in frameworks, such as PyTorch, TensorFlow or Matlab. The information of the learned network is loaded into the inference engine as a description file via ONNX. The server can operate in parallel on CPU cores, uses the integrated GPU of Beckhoff industrial PCs, or can access dedicated GPUs, such as those from Nvidia. The expected market launch is planned for Q4/22.
