Creation and conversion of ONNX
The learning process of a Machine Learning model takes place outside of the real-time, usually in a script language such as Python or R. The learned model is to be exported from the selected ML framework as an ONNX file. In order to load the description of the model in the TwinCAT runtime, the ONNX file must first be converted into a proprietary, TwinCAT 3-specific format.
An overview of the description formats for machine learning models:
Beckhoff's proprietary formats in XML and BML are directly readable from the Machine Learning Runtime. The ONNX data format must be converted to a Beckhoff proprietary format using provided converters.
Whereas ONNX and XML are openly visible formats, BML is a binary format and thus characterized above all by a small file size and an efficient loading behavior (execution time of the Configure method) in the XAR.