E-Locus - Institutional Repository of the University of Crete

Home Collections Type of Work Post-graduate theses

Post-graduate theses

Search command : Author="Ζώρας" And Author="Οδυσσέας"

Current Record: 5 of 6644

[Add to Basket]

Identifier

000464706

Title

Characterization and control of surface structures in laser-based processing using Machine/Deep Learning

Alternative Title

Χαρακτηρισμός και έλεγχος δομών επιφάνειας σε επεξεργασία με λέιζερ χρησιμοποιώντας Μηχανική/Βαθιά Μάθηση

Author

Λίτσας, Αναστάσιος Σ.

Thesis advisor

Πανταζής, Γιάννης

Reviewer

Τσαμαρδινός, Ιωάννης
Κομοδάκης, Νικόλαος

Abstract

The application of Convolutional Neural Networks (CNNs) for the characteriza- tion and control of surface structures in laser-based processing represents a ground- breaking integration of deep learning within the domain of materials science. This thesis presents the development of CNN models to analyze and extract complex spatial features from images of morphological patterns created by laser-material interactions. These patterns, critical for material properties, pose a challenge in accurately predicting laser parameters, a key factor for advancing laser process- ing technologies. In this work, we employ both Single Task Learning (STL) and Multi-Task Learning (MTL) techniques. The first part of the thesis is dedicated to STL experiments, where predictions are made for each parameter separately. In the second part, which analyzes the MTL technique, predictions of some or all target parameters are assessed. Ex- periments were conducted utilizing both synthetic and real data. STL performed exceptionally well on real data, achieving near-perfect results on synthetic data. Although improvements were observed in some predictions in the case of MTL, its overall performance did not manage to surpass STL.

Language

English

Subject

CNNs

DLIP

Laser processing

MTL

STL

Νευρωνικά δίκτυα

Issue date

2024-07-26

Collection