Post-graduate theses
Current Record: 94 of 833
|
Identifier |
000441649 |
Title |
Optimization of recyclable materials collection on conveyor belts |
Alternative Title |
Βελτιστοποίηση της συλλογής ανακυκλώσιμων υλικών από ιμάντες |
Author
|
Αγιομαυρίτη Αικατερίνη- Άρτεμις
|
Thesis advisor
|
Τραχανιάς, Παναγιώτης
|
Reviewer
|
Αργυρός, Αντώνιος
Τσακίρης, Δημήτριος
|
Abstract |
With the need for recycling of used materials growing steadily in order to save valuable
resources of our planet, and given also that the rate at which recyclable materials reach
the recycling factories and consequently the flow at which they fall on the conveyor
belts in order to separate them is particularly large, the need for more efficient ways
of separating and collecting these materials becomes apparent. In real life industrial
set-ups, a large percentage of objects pass through the conveyor belts without being
collected, at least not immediately. In addition, the main concern of a recycling factory
is profit. Based on the above, the main focus of the present work is the optimization of
the collection of the materials via the use of a robotic arm. The named optimization
is based on the capabilities of the employed robotic arm and also on the market value
of recyclable materials.
Our approach is separated into two interrelated parts, the prediction of the material
of the objects we expect to pass through the belt and their collection. In the first part,
the materials are classified into three classes (paper, plastic and aluminum) using only
information from previous throws on the belt and the characteristics of the materials
(color and size). For this part, we employ Hidden Markov Models that are capable of
accomplishing the required prediction. In the second part, a Path Planner is implemented targeting the optimization of the materials’ collection in terms of their cost.
This is implemented via a Reinforcement Learning algorithm, specifically a Q-learning
algorithm. Using a reward function the algorithm decides which is the next material
to be collected. Finally, our approach is evaluated via simulated and real results, and
its performance is also compared with that of a Proximity (Random) picker.
|
Language |
English |
Issue date |
2021-07-30 |
Collection
|
School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
|
|
Type of Work--Post-graduate theses
|
Permanent Link |
https://elocus.lib.uoc.gr//dlib/2/0/3/metadata-dlib-1628161393-60804-17039.tkl
|
Views |
561 |