A number of VEDLIoT internal use-cases were defined to demonstrate the effectiveness of the VEDLIoT toolchain and hardware platform with respect to resource efficiency, robustness, safety and security in the domains industrial IoT, automotive, and smart home.
They are described here as examples for the already wide range of ambitious requirements, that should be extended by the selected Open Call projects. Proposals submitted to VEDLIoT-Open can address similar or different areas of AIoT applications.
Representative for a large variety of industrial applications, VEDLIoT develops and optimises two different use-cases, covering aspects of Industry 4.0/5.0 topics:
Arc detection in DC distribution cabinets
- DL-based protection device for energy distribution systems
- Focusing on DC supplies and distribution setups
- Very low latency from first spark till inference, including sensing and preprocessing
- Arc localisation for faster fault detection and repair
- Main challenge: Ultra-low false negative error rate for a reliable operation
Electrical motor condition monitoring
- Battery driven, low power device with DL-based data processing
- Integrated sensing, data processing and communication capabilities
- State estimators for operational, thermal and mechanical conditions
- Robustness against external disturbing events e.g. neighboring drives
- Main challenge: usage of up to three DL models with a ultra-low energy budget
This use-case focuses on increasing the processing efficiency DL tasks over the resources that are present in the traffic environment. This will be achieved through distribution of the processing tasks over resources such as the ego vehicle, cellular base station(s) in the close proximity, as well as the cloud.
The key research areas are:
- Distributed processing including vehicle, base station and cloud
- Ensuring safety of the function, security and robustness of the distributed system
- AEB Vehicle to Pedestrian Use Case
- Increased performance using cellular base station processing resources
The Smart Mirror serves as a demonstrator for the class of smart home applications. It serves as interface between residents and the smart environment and is constructed of a combination of a mirror and a display. It visualises personalised information (e.g., news or the status of the smart home environment) and features face, object, gesture and speech recognition.
Goals within VEDLIoT:
- Virtual mirror image using 3D point clouds
- Optimising commonly used DNN models like Yolo
- Utilising embedded edge computing for resource efficiency (t.RECS)
- Combining GPU, CPU and FPGA computation
- Porting of individual parts (e.g. speech recognition as voice assistant) on minimal specialised embedded hardware (µ.RECS)