Reliable POwerDown for Industrial Drives (R-PODID-SPS)

The R-PODID project, funded by the European Union, is a cross-disciplinary initiative at the convergence of digital industry, energy, and edge AI. Its primary goal is to contribute to sustainable production, improve the application of artificial intelligence in the digital industry, and enhance the monitoring and reliability of energy systems. This project aims to provide practical solutions that address real-world challenges, fostering advancements in technology for a more sustainable future.

The overall goal of this project is to develop an automated, cloudless, short-term fault-prediction system for electric drives. Thereby, electrical and mechanical faults of machines and of the power converters driving them will become predictable within a limited prediction horizon of 12-24h.  

This will enable a power-saving shutdown of production machines during idle times, because a looming failure during the next power-on cycles can be reliably foreseen.  It will also enable reliable mitigation of dangerous faults in applications using modern power-devices like III/V-semiconductor gallium-nitride (GaN) and silicon-carbide (SiC). 

R-PODID OBJECTIVES:

• Methodology for fault-prediction model generation from sparse training sets or system simulation
• Power electronics with integrated support for embedded AI
• 24 h fault-prediction for Gallium Nitride (GaN) and Silicon Carbide (SiC) based power converters
• 24 h fault-prediction and fault mitigation for electric drives
• Sensors for reliability prediction in power modules

The partners of the project will face the challenge from different angles: investigating specific AI algorithms and how to deploy them on resource-constrained devices, developing high-power semiconductor devices based on GaN and SiC technologies that meet the requirements of modern industry and e-mobility, and developing innovative sensors and packaging solutions to collect real-time data to be used as input to the AI models and to improve the robustness of the power systems.

Supported by 33 partners, R-PODID innovations are implemented into the power modules and applied in the four use cases for conveyor belts, industrial lighting, automotive traction inverters, and a heavy-duty testbed.

TU Delft

Design of a Machine Learning based device, circuit and machine fault prediction models from sparse training data sets, specifically:

• ML prediction models for functional safety in Gallium-Nitride (GaN) and Silicon-Carbide (SiC) based power-converters (SPS group);
• Integration of self-contained, adaptive ML-based fault-prediction models into power converter modules, not relying on cloud or edge connectivity (SPS group);
• Development of novel corrosion sensor for determining the remaining useful lifetime for industrial lightning (ECTM group)
• Investigate the reliability of the packaging of GaN or SiC power drivers together with BESI (ECTM group)

Challenges

• Physics-Informed Machine Learning (PI-ML): How to use device physics and circuit models to create ML models that can be trained on sparse datasets?
• ML at the edge: How to implement ML algorithms on low-power compact devices such as microcontrollers?

ACKNOWLEDGMENT: R-PODID receives funding within the Key Digital Technologies Joint Undertaking (KDT JU) - the Public-Private Partnership for research, development and innovation under Horizon Europe – and National Authorities of Italy, Turkey, Portugal, The Netherlands, Czech Republic, Latvia, Greece, Germany, Austria, and Romania under grant agreement n° 101112338.

Disclaimer: Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the granting authority. Neither the European Union nor the granting authority can be held responsible for them.

For more information, see the project homepage.