Active Learning for Optimization of EMC processes
Funding: Scholar of the Konrad Adenauer Foundation
Contact: Youcef Hassab
Start date: 02.11.2022
With the ever increasing operating frequencies and powers, EMC has now become a major consideration on any project involving the design, construction, manufacture and installation of electrical and electronic equipment and systems. An important step in the design of components and prediction of EMC related problems is the modeling and simulation. The complexity and performance of electrical and electronic devices as well as the number and range of variables in the design spaces means that many of the Physics-Based (PB) used tools are either too slow or too inaccurate for effective design and optimization. Recently, machine learning (ML) tools and techniques have been increasingly used in the EMC domain either to improve PB approaches or to replace them. In ML, computers are used to probe vast amounts of data for structure. One requirement for the effective ML model building is the availability of these large datasets for the training and testing processes. The generation of simulation data of EMC systems using PB tools is more than often quite expensive and very time consuming.
The main goal of this project is the adaptation and extension of active learning schemes such as Bayesian Optimization (BO) to the realm of EMC. In active learning, the expensive to evaluate samples used for training and prediction are intelligently collected at each iteration to reduce the needed number of simulation runs. The active learning methods can be used for optimization tasks, modeling of systems or both at the same time. The main working points can be summarized as:
- Adaptation and foundation-laying of BO-based optimization and model building in the fields of EMC, SI, PI and bio-EMC.
- Establishment of a framework to evaluate the certainty in the solutions provided by the active learning scheme.
- Contribution to the available databases with the generated datasets for the service of the electrical engineering and EMC community.
Publications:
Data-Efficient Prediction of the Specific Absorption Rate in a Human Head Model Exposed to a Plane EM Wave Using Gaussian Process Regression Proceedings Article In: 2024 International Symposium on Electromagnetic Compatibility – EMC Europe, Bruges, Belgium, September 2-5 2024. |
Machine Learning Based Data Validation for Signal Integrity and Power Integrity Using Supervised Time Series Classification Journal Article In: IEEE Transactions on Electromagnetic Compatibility, Early Access, pp. 1 - 9, 2024. |
Generation and Application of a Very Large Dataset for Signal Integrity Via Array and Link Analysis Journal Article In: IEEE Transactions on Electromagnetic Compatibility, Early Access, pp. 1 -10, 2024. |
Application of Gaussian Process Regression for Data Efficient Prediction of PCB-based Power Delivery Network Impedance Features Proceedings Article In: 2024 IEEE 28th Workshop on Signal and Power Integrity (SPI), Lisbon, Portugal, May 12-15 2024. |
Applying Techniques of Transfer and Active Learning to Practical PCB Noise Decoupling Proceedings Article In: DesignCon 2024, Sanata Clara, USA, January 30 - February 1, 2024. |
Engineering-Informed Design Space Reduction for PCB Based Power Delivery Networks Journal Article In: IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 13, no. 10, pp. 1613 - 1623, 2023. |
Evaluation of Support Vector Machines for PCB based Power Delivery Network Classification Proceedings Article In: IEEE Conference on Electrical Performance of Electronic Packaging and Systems (EPEPS), virtual event, Austin, TX, USA, October 17-20, 2021. |
