The application of the three-dimensional pallet loading problem in an automotive factory with a metaheuristic solution approach

This study focuses on the type of pallet loading problem that can be used in both internal and external logistics disciplines, which are closely related to logistics management.The three-dimensional pallet loading problem (3D-PYP), which is the subject of this thesis, constitutes an important link in the shipping chain. In this study, boxing problems, another subsection where pallet loading or container assignment problems are encountered in the literature, are prioritized. The purpose of this study is to handle 3D-PYP, which is a derivative of truck loading problems, under the title of boxing, to produce the most suitable solution under some special constraints that the industry needs, to load the products to be shipped. Also,another purpose is the minimum number of pallets, and to produce a complete cube for boxing, with the least number of It is to be shipped using an empty box. Boxing problems are handled under the constraints of load resistance, no other objects placed on fragile box types (carton), and placing products of a single supplier's single internal route on each pallet (linked objects together). With this study, a solution proposal is presented for boxing problems and 3D-PYPs to the assembly division's internal logistics department, which is one of the largest automotive production facilities in Turkey. Currently, in this department, it is planned manually how different types of boxes will be diced and shipped in the most efficient way, based on the recognition of box types and experience. Nowadays, the development of technology and the increasing trade volume with globalization have made the optimization of logistics operations mandatory. In this study, it is aimed to provide an optimum solution with mathematical optimization tools and models to carry out these operations at high speed, high accuracy and low cost. Considering the width, length and height dimensions of the boxes used in the shipment, the 3D pallet loading problem was tried to be applied on a sample route by choosing a sample supplier, using the least empty boxes within the height limits on the standard floors used in the shipment and which is the main rule of the company. The supplier selected for this example is the supplier with the highest order volume among local suppliers. A test problem was created by targeting 90% of the box types and route diversity and constraints used. As a result of the study, 3D-PYPs solutions were presented to the company by using tabu search heuristics and genetic algorithm. In line with the results of this study, it was tried to switch from manual methods to automation in the company and to reduce the use of empty boxes and total pallets. It has been proposed to reduce the cost by increasing the full usage volume of the pallets and making efficient palletizing. Genetic algorithm and loading heuristics, objective function and constraints of the problem are defined in Python and the results are listed. Parameters of genetic algorithms in the application part of the study; box types and size information were defined as gene structure. As the population, all box types used in the selected supplier were defined. The roulette wheel selection as the selection operator was determined and applied. Thus, the chance of selecting chromosomes with high fitness values is increased. Since the box used as the crossover structure is a non-reusable constraint, the most suitable "partially mapped crossover" structure was preferred. Since the performance of genetic algorithms depends on the change of parameters, the most optimal parameters were tried to be defined in the study. In the study, the test problems were run repeatedly and parameter selection and consistency were also checked. The genetic algorithm parameters defined in the study were determined as follows. Thanks to these parameters, the study has become able to offer the solution in seconds. -Maximum number of iterations: 80 -When it iterations 25 times and does not improve, it stops finding a solution, that is, iteration. -Crossover rate: 70% -30% mutates. -Mutation rate: 0.15 After defining all the parameters and operators of this genetic algorithm, the necessary loops and commands in accordance with the general working principle of the genetic algorithm were added and the classes in Python were made ready. When the codes were run by combining the work with the Tabu Search algorithm, which is another of the main solution foundations of the study, the data results of the pallets created by GA by making random selections in each study and the occupancy rate of these pallets were obtained. Briefly the algorithm of the study, After activating the txt.file containing the initial parameters defined in the system and the box information that will serve as an input in the box.py class, the initial population is created. After the fitness values of all chromosomes (blocks and boxes) in the population are calculated, they are kept in memory according to the relevant score points. This score is evaluated as a result of scoring, and selection is made and mutation takes place according to the cross-chromosomal crossover and the defined mutation rate. Iteration limit and tabu search as maximum 80 iterations stop searching for a new solution when the limited results do not get a better result in the 25th iteration. After all these definitions were made, the model was run in python. Solution results were obtained according to the number of boxes and the solution pool iteration probability between 0.16 and 0.1 seconds on average for each pallet. In the conclusion and evaluation part of the study, the solution results provided by the modeling in python and the solution results made manually in the company are compared. One of the most important findings in this study is that since the operation was transferred from manual to digital, the results were made independent of being dependent on the person and experience. This provided flexibility in operation as well as increased productivity. As a result of the comparison of inexperienced operator, experienced operator and system results, it is presented that cost and labor savings are provided to the company and the processing time is reduced. When the modeling is run, the problem of placing 8 different box types and 46 boxes belonging to 1 route of a single supplier with maximum efficiency has been fit into a total of 5 pallets with this study. 4 of these pallets have a 100% filling volume and 1 have a 32% filling volume. The reason why the final pallet efficiency is like this is because the boxes to be placed are finished. All these results were able to provide a solution in an average of 0.10 seconds when run at once. By eliminating the boxes used and rerunning the program separately, the solution was reached in 0.42 seconds in total. On the other hand, placing all boxes for this sample size takes 32 minutes, including the preparation time of the data manually. Within the scope of this study, the same problem was solved separately by a 6-year-experienced employee and a 1-year-experienced employee. In this case, the 6-year-experienced employee was able to fit all the boxes on 6 pallets in total, and the 1-year-experienced employee was able to fit 7 pallets in total. In addition to the increase in the number of pallets, an increase in the use of empty boxes has emerged in the palletizing solution made by the experienced member and the inexperienced member, according to the system. As can be seen from the results of the comparison, this study presented the company with an improvement in terms of reducing 1 operator due to the reduction of handling operations on the supplier side and the efficiency of shipping more boxes with fewer pallets. In addition, it is planned to submit proposals for the implementation of the improvement in all company chains as a global kaizen. The general contribution and difference of the study to the literature is that, apart from the fast and highly efficient solution it offers to a real life problem with such large masses, it is generally focused on the external logistics departments in case of boxing or container problems, while it includes both the manufacturer and distributor principles in which the box distribution is made, such as internal logistics. It has been shown that solutions can be proposed within the chapters and can be considered as a study subject. In addition, it is possible to integrate the proposed GA for use in robotics in unmanned and contactless automatic palletizing projects, which is one of the company's future projects to be carried out within the scope of Industry 4.0.