June 19, 2026 | 11:30 AM – 1:00 PM
Department of Electronics, Information and Bioengineering - Politecnico di Milano
Emilio Gatti Conference Room (Building 20)
Speaker Presentations: Sofia Guerra, Matteo Moscatelli, Federico Porcari
Speaker Pitch: Guoming Shi, Giuseppe Bavaresco, Gabriele Bianchi
Contacts: phd-inf@polimi.it
Abstract
IT Talks is a seminar series dedicated to Ph.D. students and researchers in the field of Information Technology, conceived as a space for discussion on cutting-edge research, advanced methodologies, and interdisciplinary applications.
The fifth seminar will take place on June 19 2026, from 11:30 AM to 1:00 PM, in the Emilio Gatti Conference Room (Building 20). The event will feature three invited speakers and three speaker pitch.
Sofia Guerra
Vision-Based Robust Closed-Loop Control for Quality Optimization in Laser Cutting
Laser cutting is widely adopted in smart manufacturing thanks to its precision, speed, and flexibility. However, maintaining stable process conditions and high cut quality remains challenging due to process instabilities, defect formation, and variability across operating conditions. This PhD research focuses on data-driven and learning-based approaches for virtual sensing, process monitoring, and closed-loop regulation in laser cutting systems.
The work investigates coaxial and external vision-based monitoring strategies to extract process-relevant features in real time, enabling defect detection and adaptive process control. A major contribution concerns the development of hierarchical closed-loop control architectures for the minimization of both single and multiple defect types through dynamic regulation of process parameters. In particular, coaxial monitoring is exploited for melt-pool analysis, process state classification, and supervisory control of complex cutting trajectories.
To improve industrial applicability, the research also addresses robustness and transferability through domain adaptation strategies, contamination analysis, and auxiliary functionalities such as automatic nozzle centering based on NIR coaxial vision. The overall goal is to enhance the robustness and flexibility of laser cutting systems, supporting the transition toward intelligent and autonomous manufacturing environments.
Matteo Moscatelli
Data-Driven, Learning-Based Approaches for Virtual Sensing and Closed-Loop Regulation of Laser Welding for Smart Manufacturing
Laser welding is increasingly adopted in smart manufacturing due to its precision, speed and flexibility, yet accurate path execution remains challenging when dealing with complex seam geometries, positioning errors and part-to-part variability. This PhD research focuses on data-driven and learning-based approaches for virtual sensing, seam tracking and closed-loop trajectory regulation in robotic laser welding.
The work investigates vision-based methods to estimate the seam position and relevant spatial features in real time, enabling the system to correct the welding trajectory during execution. In parallel, autonomous trajectory generation strategies are explored to reduce manual programming effort and increase adaptability to different workpieces and seam geometries. The final goal is to improve accuracy and robustness in robotic laser welding, supporting the transition toward more flexible and intelligent manufacturing systems.
Federico Porcari
Explainable AI for Data-Driven Control: An Inverse Optimal Control Approach
As data-driven models become increasingly common in control applications, their lack of transparency raises important questions about trust and reliability. This presentation provides a concise outlook on the explainability framework, introducing explanations as a two-step process: first simplifying an opaque model into a useful explanatory representation, and then extracting human-interpretable information from it.
The main focus is on the specific challenges of explainability in control systems, where classical XAI methods, typically designed for static, open-loop models, may fail to capture the closed-loop behaviour that actually determines controller performance. To address this limitation, the talk presents a control-theoretic approach based on inverse optimal control. By observing closed-loop trajectories generated by an unknown controller, the proposed method interprets the controller's behavior through the lens of an equivalent LQR problem.