This reference design features an analog front end for DC arc detection in solar applications based on artificial intelligence (AI). DC arcing induces a high-frequency noise onto the DC sting current. To detect the arcing frequency, data is acquired and feed into an embedded AI model, which is trained to identify the arc. Compared to traditional arc detection approaches this achieves higher accuracy with less computational efforts. Besides the signal-chain for arc detection this design offers features for collecting and labeling arcing data for embedded AI model training. This hardware works with TMDSCNCD28P55X, controlCARD of C2000™ F28P55xdevices, as well as other C2000 controlCARDs in 180-pin connector and is part of a AI based arc detection tool chain, including different software tools for collecting arc data, training an embedded AI model and validating the system.
Features
- 4-channel analog front end for AI-based arc detection
- Configurable analog front end with band-pass and notch filter
- String voltage and arc gap voltage measurement inputs for training data acquisition
- Auto-labeling circuits to generate labeled arcing data
- Works with TMDSCNCD28P55X, controlCARD of C2000 F28P55x devices, as well as other C2000 controlCARDs in 180-pin connector
- Selected embedded AI models for a quick start into AI-based arc detection