7th NorCal Control Workshop - 4 posters from MESA Lab
Website: https://sites.google.com/ucdavis.edu/7th-norcal-control-workshop [4/25/2025]
[Last year 2024 was in UC Merced https://sites.google.com/view/6thnorcalcontrolworkshop]
Authors/Title/Presenter/Abstract:
Rafal Krzysiak, Sachin Giri, Derek Hollenbeck, YangQuan Chen. “Modeling and control of prescribed fire with UAVs as Sensors and Actuators”. Poster at Friday, April 25, 2025, 7th NorCal Control Workshop, University of California, Davis. Presenter: Rafal Krzysiak (3rd year ME Ph.D. student). Abstract: Prescribed fires are critical management tools for maintaining ecosystem health, reducing wildfire risk, and promoting biodiversity. However, effective execution of prescribed burns poses significant challenges due to inherent complexities in fire behavior and the need for precise control and monitoring. This work presents a comprehensive framework for the modeling and control of prescribed fires utilizing Unmanned Aerial Vehicles (UAVs) as integrated sensing and actuation platforms. We propose a hybrid approach combining physics-based fire propagation models with real-time data assimilation from UAV-based sensors, including multispectral and thermal imaging. The digital twin approach enables accurate predictions of fire spread dynamics, facilitating adaptive control strategies. By leveraging UAVs which can autonomously respond to changing environmental conditions, ensuring targeted and safe fire management, improvements can be seen in precision, responsiveness, and safety compared to traditional prescribed fire methods. This research paves the way toward autonomous, intelligent fire management systems, significantly enhancing land management practices and wildfire mitigation strategies. [poster pdf]
Osama Fuad Abdel Aal, Sinan Ozbek, YangQuan Chen. “Controlled Optimization Processes by Systematic Designs: A Dissipativity-Based Approach” Poster at Friday, April 25, 2025, 7th NorCal Control Workshop, University of California, Davis. Presenter: Osama Fuad Abdel Aal (3rd year ME Ph.D. student). Abstract: Finite-time and fixed-time optimization are crucial in control and decision-making systems that demand fast, guaranteed convergence. Unlike traditional methods that only ensure asymptotic convergence, these approaches are designed to reach the optimal solution within a finite or uniformly bounded time, regardless of initial conditions (in the fixed-time case). This is especially important in real-time, safety-critical, or resource-constrained applications like robotics, autonomous systems, and networked control, where delays or prolonged computation
can compromise performance or safety. This poster addresses controlled optimization processes with a focus on convergence properties considering various proposed algorithms with a unified framework. Toward this goal, a new analysis of finite-, fixed-, and prescribed-time convergent optimization algorithms is presented in the perspective of dissipativity theory. This perspective enables the unification of time-constrained optimization algorithms under the framework of dissipativity control theory, and may enable the design of new algorithms that satisfy theses convergence properties. (poster pdf)
Mohammad Partohaghighi, Roummel Marcia, YangQuan Chen. “Tail-index informed federated learning from long tailed dataset.” Poster at Friday, April 25, 2025, 7th NorCal Control Workshop, University of California, Davis. Presenter: Mohammad Partohaghighi (2nd year EECS Ph.D. student). Abstract: Federated Averaging (FedAvg) is a key algorithm in Federated Learning (FL), delivering strong results by combining locally trained models from distributed, often diverse datasets. Yet, its performance heavily depends on hyperparameters like learning rate, batch size, local epochs, and the proportion of clients involved in each communication round. In this paper, we introduce a structured approach to fine-tune these hyperparameters, achieving notable improvements in convergence speed and final accuracy on a long-tailed MNIST benchmark. Through sequential searches for each parameter—learning rate, batch size, local epochs—and then the fraction of participating clients, we illustrate their individual roles in enhancing performance under class-imbalanced scenarios. Our findings show that thoughtfully selected hyperparameters can significantly outperform default configurations, offering valuable guidance for real-world FL implementations. (poster pdf)
Derek Hollenbeck, Sachin Giri, YangQuan Chen. “Empirical observability-based source term estimation using mobile sensor trajectories.” Poster at Friday, April 25, 2025, 7th NorCal Control Workshop, University of California, Davis. Presenter: Derek Hollenbeck (postdoc research fellow, manager MESA Lab and CMERI). Abstract: Real time parameter estimation relies on a fast running model of the system and a method for optimizing the parameters. The source term estimation problem of a trace atmospheric gas, is typically represented as a PDE with high-dimensional order -- making it very difficult to solve in a real-time fashion. By utilizing the empirical Gramian, a fast running low fidelity surrogate model, and a high fidelity digital twin of the emission source, the optimal trajectory can be explored, in the observability sense, to estimate the parameters of the system. (poster pdf)
Created and last updated by Prof. YangQuan Chen 4/24/2025.
