Physics-informed machine learning-enabled digital twin implementation for power electronic converters

Kerry Sado

Sebastian Oviedo

Outline

• What are Digital Twins (DTs)?
• Applications of DTs.
• Challenges.
• AI in Power Electronics.
• Integration of PIML with DTs.

​

​

What are Digital Twins (DTs)?

• A digital twin is a virtual model of physical assets and processes and, in essence, is a computer program that uses real-world data to create simulations.
• In practice, a digital twin is based on sensors that gather real-time data and connect to a cloud-based system that receives and processes it all.
• Digital twin technology can be traced back to the 1960s, when NASA needed to simulate systems in space.

[1] Grieves, Michael. (2019). Virtually Intelligent Product Systems: Digital and Physical Twins.

What are Digital Twins (DTs)?

• Unlike static data models, digital twins are dynamic, “living” entities that evolve in real time.
• They learn, update, and communicate with their physical twins (PT) by exchanging data throughout the PT’s lifecycle using AI, machine learning, and IoT technologies.

[1] Grieves, Michael. (2019). Virtually Intelligent Product Systems: Digital and Physical Twins.

DT Applications

[2] Guo, J., Lv, Z. Application of Digital Twins in multiple fields. Multimed Tools Appl.

Challenges of DTs.

• Data acquisition and integration.
• Scalability and complexity.
• Model accuracy and validation.

​

Challenges of DTs.

• Physical model.

Circuit topology of dc–dc Buck converter �

Challenges of DTs.

• Mathematical model.

[3] Y. Peng, S. Zhao, and H. Wang, “A digital twin based estimation method for health indicators of dc–dc converters,” IEEE Transactions on Power Electronics, vol. 36, no. 2, 2021.

AI in Power Electronics

A novel approach to the design, analysis, and modeling of power systems:

[4]:  Number of technical papers on using artificial intelligence in power electronics published annually from 1990 to May 2020. (Graph: IEEE Transactions on Power Electronics)

Data-Driven Neural Network Constraints

[5] General Structure of a Feed Forward Neural Network: Johnson, J., et al. Artificial neural networks improve early outcome prediction and risk classification in out-of-hospital cardiac arrest patients admitted to intensive care (2020) 

Integration of Physics-Informed ML with DTs

[6] Structure of NN that considers physical constraints: Yanan, G. et al. Solving Partial Differential Equations Using Deep Learning and Physical Constraints (2020)

Experimental testbed

• To validate the DT model, a single converter will be used.
• The model will be deployed on a National Instruments (NI) compact RIO (cRIO).
• The cRIO chassis will act as the interface between the physical converter and the DT model, using Ethernet for communication.
• The power converter will be implemented using the Imperix PEB8038 - Half bridge SiC power module with optical gate drive circuits.
• Nested loop controls integrated into the Imperix BBox control platform will drive the converter.
• An electronic DC load will be used to apply constant and pulsed loads to the system.

Simulations

• A buck 200/100 V converter was simulated to generate data.
• The PINN's architecture consists of 6 fully connected layers, each with 50 feed-forward neurons

Thank you!