| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

Show document Help

Title:Razvoj premestitvenih strategij v večglobinskem regalnem skladiščnem sistemu avtomatskih vozičkov : doktorska disertacija
Authors:ID Marolt, Jakob (Author)
ID Lerher, Tone (Mentor) More about this mentor... New window
Files:.pdf doktorat_7.0_FINAL_brez_praznih_strani.pdf (8,27 MB)
MD5: 9124D6D776A6B806F9F707C1FC3B3B87
 
Language:Slovenian
Work type:Doctoral dissertation
Typology:2.08 - Doctoral Dissertation
Organization:FS - Faculty of Mechanical Engineering
Abstract:V doktorski disertaciji je obravnavan premestitveni problem transportno skladiščnih enot (TSE) v večglobinskem skladiščnem sistemu avtomatskih vozičkov (AVS/RS). Pri procesu odpreme TSE lahko zaradi več globin prednja TSE ovira dostop do ciljne TSE. Na podlagi uporabe učinkovite skladiščne in premestitvene strategije lahko ovirajočim TSE izberemo skladiščno mesto, kjer minimiramo povprečni čas enojnega in dvojnega delovnega cikla (DDC) in posledično povečamo pretočno zmogljivost večglobinskega AVS/RS skladiščnega sistema. V kontekstu problematike stohastične odpreme TSE, kjer vrstni red odpreme TSE ni vnaprej poznan, smo z uporabo metode Markovskih verig za kombinacijo skladiščne strategije najglobljega mesta (DF) in premestitvene strategije DF izpeljali in predstavili analitični model. Zaradi izrazite kompleksnosti analitičnega modela smo prav tako razvili in predstavili empirični model, ki temelji na linearnih in kvadratnih regresijskih enačbah . Oba predlagana modela smo verificirali s pomočjo simulacijskega modela diskretnih dogodkov (DES). Analiza rezultatov je pokazala visoko ujemanje napovedi povprečnih časov DDC obeh predlaganih modelov z rezultati simulacije. V okviru doktorske disertacije smo analizirali več različnih kombinacij skladiščnih in premestitvenih strategij. Namen te analize je bila identifikacija najučinkovitejše kombinacije skladiščne in premestitvene strategije za zmanjšanje povprečnega časa DDC in povečanje pretočne zmogljivosti večglobinskih AVS/RS skladiščnih sistemov. Pri simulacijski analizi DES smo spreminjali število globin in število stolpcev skladiščnega regala (skladiščnih mest vzdolž regalnega hodnika) pri čemer je skupno število skladiščnih mest (zalogovna velikost skladišča) ostalo enako pri vseh konfiguracijah skladišča. Analiza je pokazala, da skladiščni sistemi z večjim številom globin skladiščnega regala omogočajo doseganje krajših povprečnih časov DDC. Prav tako smo ugotovili, da kombinacija skladiščne strategije DF in premestitvene strategije najbližjega soseda (NN) velja za najbolj učinkovito, še posebej pri uporabi AVS/RS skladiščnih sistemih s petimi in šestimi globinami. Problem deterministične odpreme TSE, kjer je vrstni red odpreme TSE poznan v naprej, smo reševali z algoritmi spodbudnega učenja. Analizirali smo učinkovitost algoritmov Deep Q-Network (DQN) in Proximal Policy Optimization (PPO). Razvili in predstavili smo štiri načine zapisa stanja večglobinskega AVS/RS skladiščnega sistema in analizirali, s katerim zapisom stanja so se agenti najučinkovitejše priučili reševanja premestitvenega problema TSE . Rezultate algoritmov DQN in PPO smo primerjali z rezultati celoštevilčnega programa in ugotovili, da so agenti priučeni z algoritmom PPO reševali premestitveni problem z manjšim povprečnim številom premestitev. Nadaljnja analiza rezultatov je pokazala, da so bili agenti priučeni z algoritmom DQN bolj zanesljivi pri reševanju premestitvenega problema TSE, v primerjavi z agenti priučenimi z uporabo algoritma PPO. V primerjalni analizi smo ugotovili, da lahko priučeni agenti rešujejo premestitveni problem TSE skoraj optimalno.
Keywords:Intralogistika, večglobinski AVS/RS skladiščni sistem, premestitveni problem TSE, analitično in numerično modeliranje, spodbudno učenje, analiza učinkovitosti
Place of publishing:Maribor
Place of performance:Maribor
Publisher:[J. Marolt]
Year of publishing:2024
Number of pages:XVII, 135 str., [8] str. pril.
PID:20.500.12556/DKUM-87126 New window
UDC:[519.21+519.87]:621.796.5-52(043.3)
COBISS.SI-ID:221252099 New window
Publication date in DKUM:16.12.2024
Views:0
Downloads:12
Metadata:XML DC-XML DC-RDF
Categories:KTFMB - FS
:
Copy citation
  
Average score:(0 votes)
Your score:Voting is allowed only for logged in users.
Share:Bookmark and Share


Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Licences

License:CC BY 4.0, Creative Commons Attribution 4.0 International
Link:http://creativecommons.org/licenses/by/4.0/
Description:This is the standard Creative Commons license that gives others maximum freedom to do what they want with the work as long as they credit the author.
Licensing start date:21.02.2024

Secondary language

Language:English
Title:Development of relocation strategies in a multi-deep automated vehicle storage and retrieval system
Abstract:In the doctoral dissertation, the relocation problem of Stock Keeping Units (SKUs) in a multi-depth Automated Vehicle Storage and Retrieval System (AVS/RS) is addressed. During the dispatch process of SKUs, the presence of multiple depths may cause a front SKU to obstruct access to the target SKU intended for dispatch. By applying an efficient storage and relocation strategy, obstructing SKUs can be relocated to a storage location that minimizes the average time of both single and double command cycles (DCC), thereby enhancing the SKU throughput performance of the multi-depth AVS/RS storage system. In the context of stochastic dispatch of SKUs, where the order of SKU dispatch is not known in advance, we derived and presented an analytical model using the Markov chain method for the combination of the depth-first (DF) storage and DF relocation strategies. Given the significant complexity of the analytical model, we also developed and presented an empirical model based on linear and quadratic regression equations. Both proposed models were verified using a discrete event simulation (DES) model. The analysis of the results demonstrated a high correlation between the predictions of average DCC times by both proposed models and the simulation outcomes. Furthermore, several different combinations of storage and relocation strategies were analyzed. The purpose of this analysis was to identify the most effective combination of storage and relocation strategies for reducing the average DCC time and increasing the throughput capacity of multi-depth AVS/RS storage systems. In the DES simulation analysis, we varied the number of depths and the number of columns (storage locations along the rack aisle) while keeping the total number of storage locations the same across all configurations. The analysis showed that AVS/RS with a greater number of depths enabled shorter average DCC times. Additionally, it was found that the combination of the DF storage strategy and the nearest neighbor (NN) relocation strategy was the most effective, particularly in AVS/RS storage systems with five and six depths. In addressing the relocation problem with deterministic SKU dispatch, where the order of SKU dispatch is known in advance, reinforcement learning algorithms were employed. The efficiency of the Deep Q-Network (DQN) and Proximal Policy Optimization (PPO) algorithms was analyzed. Four methods of representing the state of the multi-depth AVS/RS storage system were developed and assessed to determine which state representation was most effective for training agents to solve the SKU relocation problem. The analysis revealed that, compared to DQN and an integer program, agents trained with the PPO algorithm required a lower average number of relocations to solve the SKU relocation problem. Further analysis of the results indicated that agents trained with the DQN algorithm were more reliable in solving the SKU relocation problem compared to those trained with the PPO algorithm. Comparative analysis has shown that agents, when trained using reinforcement learning algorithms, could solve the SKU relocation problem nearly optimally.
Keywords:Intralogistics, multi-deep AVS/RS storage system, relocation problem of SKU, analytical and numerical modeling, reinforcement learning, efficency analysis


Comments

Leave comment

You must log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back
Logos of partners University of Maribor University of Ljubljana University of Primorska University of Nova Gorica