Project Mission and Main Objectives
R-PODID is an ambitious three-year research and innovation project aimed at developing automated, cloudless, short-term fault prediction for electric drives, power modules, and power devices. The project's core mission is to enable predictable electrical and mechanical faults within a limited prediction horizon of 12-24 hours, fundamentally transforming how industrial power systems operate.
The project addresses five primary objectives: developing methodology for fault-prediction model generation from sparse training sets, creating power electronics with integrated embedded AI support, implementing 24-hour fault prediction for Gallium-Nitride (GaN) and Silicon-Carbide (SiC) based power converters, extending fault prediction capabilities to electric drives, and optimizing sensors for reliability prediction in power modules. These objectives collectively target energy-saving, cost reduction, and CO2 emission reduction by enabling safe shutdown of idle production machines while ensuring reliable restart capabilities.
Core Technologies and Technical Innovation
R-PODID employs three main technological innovations that distinguish it from existing solutions. First, the project integrates self-contained, adaptive fault-prediction directly into power converter modules without requiring cloud or edge connectivity, ensuring data sovereignty and reducing communication bandwidth requirements. Second, it develops advanced model generation methods for AI-based device, circuit, and machine fault prediction using sparse training datasets, addressing the fundamental challenge of limited fault occurrence data. Third, the project focuses on lifetime improvement and AI assistance for functional safety in next-generation GaN and SiC-based power converters.
The technical architecture combines hybrid AI models that augment compact mathematical models with artificial neural networks, employing advanced techniques including Generative Adversarial Networks (GANs) for data augmentation and temporal convolutional neural networks with self-attention mechanisms. The system utilizes a custom RISC-V-based computing platform with domain-specific hardware accelerators optimized for real-time AI inference within power conversion systems.
Integration and Compatibility Framework
R-PODID technology integrates seamlessly into existing industrial power converters through modular hardware and firmware building blocks based on open-source RISC-V architecture. The system supports various power converter topologies and machine types by developing generalized fault-prediction methodologies that can be adapted to specific applications while maintaining compatibility across different electric drive configurations.
The project implements specialized sensor integration strategies, including advanced Hall-effect current sensors with 10-MHz bandwidth and 300-A input range, alongside optimized shunt-based current sensors with improved temperature compensation. These sensors are designed for direct integration into power modules, providing the necessary data streams for local fault prediction without external dependencies.
Milestones and Expected Impact
Key performance milestones include achieving 85% prediction accuracy with 50% reduction in required training data, reducing model memory footprint by 80%, and maintaining computational complexity reduction of 85% compared to conventional approaches. The project also targets real-time fault detection within one minute with over 90% accuracy for various failure modes.
R-PODID's societal impact extends across commercial, environmental, scientific, and social dimensions. The project targets significant CO2 emission reductions, with conservative estimates suggesting potential savings of 14 million tons of CO2 emissions if implemented in just 20% of industrial drives. An optimistic scenario reaching 50% market penetration could contribute to approximately 1-2% reduction of overall EU-27 CO2 emissions.
Scaling Barriers and Industry Challenges
The primary barriers to widespread R-PODID adoption across industries include the complexity of integrating AI-enabled sensors into existing power infrastructure, the need for industry-specific model training and validation, and the requirement for specialized expertise in both power electronics and embedded AI systems. Different industries face unique challenges, e.g., automotive applications demand extremely high reliability standards and extensive validation procedures, while food processing requires compliance with hygiene and safety regulations, and semiconductor manufacturing needs ultra-precise control and minimal downtime tolerance.
Economic considerations also present scaling challenges, as the technology requires initial investment in upgraded power modules and sensor integration, though these costs are offset by reduced maintenance expenses and energy savings over the system lifecycle. R-PODID addresses these barriers through open-source development approaches and comprehensive validation procedures designed to demonstrate commercial viability across diverse industrial applications.
Project link: