Browsing by Author "Bulut, Önder"

Now showing 1 - 20 of 32

A Genetic Algorithm for a Real-Life Scheduling Problem in the Valve Industry
(Springer Science and Business Media Deutschland GmbH, 2023) Gamze Esma Bektaş; Ege Cömert; Ezgi Sena Yılmaz; Melis Tan Tacoglu; Önder Bulut; Burçin Kasap; Pınar Aydın; Mustafa İnceoğlu; Eda Badak; Hasan Şenol; Yılmaz, Ezgi Sena; Taçoğlu, Melis; Bulut, Önder; Bektaş, Gamze Esma; Cömert, Ege; Kasap, Burçin; Özcureci, Kaan; N.M. Durakbasa , M.G. Gençyılmaz
In this paper a real-life flexible job shop scheduling problem (FJSSP) for valve production having sequence-dependent setup times and machine unavailability constraints is studied. The aim is to minimize the weighted sum of earliness and tardiness of the scheduled jobs. Since this problem is known to be in the NP-Hard class we develop a Genetic Algorithm (GA) enriched with Iterated Local Search (ILS) to obtain near-optimal solutions for both the company’s problem and also for much larger instances in a reasonable run time. For the real-life implementation we develop a user-friendly Decision Support System (DSS) consisting of databases for the inputs a GA algorithm embedded as Python code and a Gantt Chart representation of solutions as the output. © 2023 Elsevier B.V. All rights reserved.
A Lost Sales Make-to-Stock System with Batch Demand and Batch Production
(Springer Science and Business Media Deutschland GmbH, 2024) Sinem Özkan; Mert Yandımata; Önder Bulut; Yandımata, Mert; Özkan, Sinem; Bulut, Önder; N.M. Durakbasa , M.G. Gençyılmaz
This study considers a production setting of a single item single production resource make-to-stock production system with batch demand batch production and lost sales. Demand arrives as a Poisson process with a randomly distributed batch size. It is assumed that the batch demand can be partially satisfied and unsatisfied demands are lost. Production time follows an exponential distribution and each production order contains a single lot that does not exceed the maximum lot size. The decision to be considered in managing such a system is the production control decision which determines when to start production and how much to produce in a batch. This study contributes to the technical literature on controlling make-to-stock systems by addressing a single production resource simultaneously with batch demand and batch production. The system is formulated with the dynamic programming and the model is solved using the value iteration algorithm. In the light of the results obtained by using this algorithm it has been shown by numerical studies that the optimal production policy is a dynamic policy. © 2024 Elsevier B.V. All rights reserved.
Citation - Scopus: 1
A Simulation Based Analysis for Sewing Lines of Apparel Industry
(Springer Science and Business Media Deutschland GmbH, 2023) Berk Kaya; Zeynep Hazal Soyan; Işılay Aydın; Sanemnaz Yurteri; Zilan Gerilakan; Önder Bulut; Burak Özdeş; Can Elhan; Zeynep Rala; Şafak Birol; Birol, Şafak; Yurteri, Sanemnaz; Kaya, Berk; Aydın, Işılay; Bulut, Önder; Gerilakan, Zilan; Soyan, Zeynep Hazal; N.M. Durakbasa , M.G. Gençyılmaz
One of the frequently encountered problems in the labour-intensive apparel industry is the deviation from the target deadlines. The biggest observed reason for this problem is that the standard operation times and therefore production targets are being defined as deterministic values in the stochastic environment. The statistical studies carried out in this project show that the standard deviation of the operation times is too significant to be ignored. The natural randomness in the performance of the workers and additional factors such as fatigue and learning effect are some of the most important reasons that increase the variability in the system. The project is carried out in conjunction with one of Turkey’s top companies within the apparel sector TYH Tekstil A.Ş. Different policies are applied by the company to dynamically solve the problems encountered during production. In the literature review different simulation-based policy methods for a sample order were examined. However since the company’s sewing line has its unique characteristics the reviewed policies alone were determined not to be suitable for use. This project aims to determine the mentioned randomness parameters by analyzing the day-hour variability of the operation times statistically and provide the company with a tool that will simulate the sewing lines ahead of production using the order-specific inputs and data analysis dynamically analyze and improve the performance of the sewing lines by Tabu search algorithm and visually/verbally report any problematic operations/occasions that may arise and performance improvements made by the decision policies for minimizing deviations from target deadlines. To increase the flexibility of the system and make it dynamic the Python programming language was used. © 2023 Elsevier B.V. All rights reserved.
Citation - WoS: 12
Citation - Scopus: 11
An efficient procedure for optimal maintenance intervention in partially observable multi-component systems
(Elsevier Ltd, 2024) Oktay Karabağ; Önder Bulut; Ayhan Özgür Toy; Mehmet Murat Fadiloglu; Toy, Ayhan Özgür; Karabağ, Oktay; Bulut, Önder; Fadıloğlu, Mehmet Murat
With rapid advances in technology many systems are becoming more complex including ever-increasing numbers of components that are prone to failure. In most cases it may not be feasible from a technical or economic standpoint to dedicate a sensor for each individual component to gauge its wear and tear. To make sure that these systems that may require large capitals are economically maintained one should provide maintenance in a way that responds to captured sensor observations. This gives rise to condition-based maintenance in partially observable multi-component systems. In this study we propose a novel methodology to manage maintenance interventions as well as spare part quantity decisions for such systems. Our methodology is based on reducing the state space of the multi-component system and optimizing the resulting reduced-state Markov decision process via a linear programming approach. This methodology is highly scalable and capable of solving large problems that cannot be approached with the previously existing solution procedures. © 2024 Elsevier B.V. All rights reserved.
Citation - WoS: 5
Citation - Scopus: 5
An embedded Markov chain approach to stock rationing under batch orders
(ELSEVIER SCIENCE BV, 2019) Mehmet Murat Fadiloglu; Onder Bulut; Fadıloğlu, Mehmet Murat; Bulut, Önder
We provide an extension to the embedded Markov chain approach of Fadiloglu and Bulut (2010) for the analysis of lot-per-lot inventory systems with backorders under rationing. This extension generalizes the method so that it can be used for the analysis under (Q R) policy. We introduce a new embedded Markov chain of higher dimensionality that keeps track of the ordering process. We finally provide a recursive procedure to generate the steady-state probabilities from the chain and obtain the steady-state distribution. (C) 2018 Elsevier B.V. All rights reserved.
Analysis of a Turkey Meat Production Agri-Chain: A Simulation Study
(Springer Science and Business Media Deutschland GmbH, 2024) Nazlı Karatas Aygün; Levent Kandiller; Önder Bulut; Aygün, Nazli Karataş; Kandiller, Levent; Bulut, Önder; N.M. Durakbasa , M.G. Gençyılmaz
Poultry meat will take the greatest share of global meat consumption within ten years because of the ongoing increase. It is anticipated that this upward trend in consumption will result in a rise in the amount of poultry meat produced. Poultry meat agri-chains are composed of various interdependent components including a hatchery many farms and coops and a slaughterhouse generally. The current study tackles the analysis of a real-life agri-chain reaction to changes in hatchery slaughterhouse and farm/coop capacities under different stochastic demand parameters. A discrete-event simulation model is used. The simulation model introduced here is a generic model such that model parameters can be played to reveal different real-life cases. After verifying the model using a toy problem the model is used to observe the impacts of several model parameters on the system-wide performance measures. © 2024 Elsevier B.V. All rights reserved.
Analysis of make-to-stock queues with general processing times and start-\rup and lost sales costs
(2022) Önder Bulut; Sinem Özkan; Özkan, Sinem; Bulut, Önder
We consider a make-to-stock environment with a single production unit that\rcorresponds to a single machine or a line. Production and hence inventory are\rcontrolled by the two-critical-number policy. Production times are independent\rand identically distributed general random variables and demands are generated\raccording to a stationary Poisson process. We model this production-inventory\rsystem as an M/G/1 make-to-stock queue. The main contribution of the study is to\rextend the control of make-to-stock literature by considering general production\rtimes lost sales and fixed production costs at the same time. We characterize the\rlong-run behaviour of the system and also propose a simple but very effective\rapproximation to calculate the control parameters of the two-critical-number\rpolicy. An extensive numerical study exhibits the effects of the production time\rdistribution and the system parameters on the policy control levels and average\rsystem cost.
Citation - WoS: 2
Citation - Scopus: 4
Analysis of make-to-stock queues with general processing times and start-up and lost sales costs
(Balikesir University, 2022) Sinem Özkan; Önder Bulut; Özkan, Sinem; Bulut, Önder
We consider a make-to-stock environment with a single production unit that corresponds to a single machine or a line. Production and hence inventory are controlled by the two-critical-number policy. Production times are independent and identically distributed general random variables and demands are generated according to a stationary Poisson process. We model this production-inventory system as an M/G/1 make-to-stock queue. The main contribution of the study is to extend the control of make-to-stock literature by considering general production times lost sales and fixed production costs at the same time. We characterize the long-run behaviour of the system and also propose a simple but very effective approximation to calculate the control parameters of the two-critical-number policy. An extensive numerical study exhibits the effects of the production time distribution and the system parameters on the policy control levels and average system cost. © 2023 Elsevier B.V. All rights reserved.
Bus Spare Parts Demand Forecasting via Holt Winter’s Method and Support Vector Regression Algorithm
(Springer Science and Business Media Deutschland GmbH, 2025) İlker Mutlu; Önder Bulut; Mutlu, İlker; Bulut, Önder; N.M. Durakbasa , K.G. Gülen
Demand forecasting and spare parts availability are crucial in managing spare parts for public transport systems. The condition of public transport buses is critical for service quality. Regular maintenance and repair of buses can increase public transport usage. This case study focuses on forecasting spare parts demand for a public transport company with more than 1700 buses and five workshops. Considering the variety of bus models and wide range of parts effective demand forecasting is essential to avoid unnecessary holding costs or service interruptions. This study aims to apply two different demand forecasting methods Winter’s Method and Support Vector Regression to a public transport company that makes forecast demands based on experience and to interpret the results. While applying the Support Vector Regression method demand dates and bus ages were used as factors. Parameter optimization and demand forecasting methods were applied to the demand data using Weka software. Data were collected from the company's ERP software between 2019 and 2024. Spare parts were determined from five spare parts from different workshop units of the company. As a result of the study MAPE values were calculated. According to the calculated MAPE values it was determined that both methods gave similar results and successfully captured the demand structure. © 2025 Elsevier B.V. All rights reserved.
Control and Performance Analysis of Three Station Make-to-Stock Production Lines
(2021) Yücel, Özgün; Bulut, Önder
Üretim sistemleri üzerine yapılan çalışmalar, on yıllardır rastgelelik, müşteri gereksinimleri, üretim süreçlerinin belirli özellikleri ve sistem maliyetleri ile başa çıkmaktadır. Stoğa üretim, müşteri hizmet düzeyini artırıp üretim, envanter ve kıtlıklarla ilişkili maliyetlerin dengelenmesini kolaylaştırır. Bu çalışma, stoğa üretim ortamında seri olarak düzenlenmiş üç istasyondan oluşan üretim sistemlerinin üretim kontrolü ve performans değerlendirmesini ele almaktadır. İlk olarak, tek makineli istasyonlar, istasyonlar arasında yer alan yarı mamül stokları ve bitmiş ürün stoğu içeren üretim sistemlerinin eniyi kontrol problemleri incelenmiştir. Taleplerin bir Poisson sürecine göre geldiği bu çalışmada, son ürün stoğundan anında karşılanamayan talepler için kayıp satış bedeli ödenir. Üstel olarak dağıtılmış işlem sürelerine sahip makinelerden oluşan sistem ana model olarak tanımlanırken, genişletilmiş modellerde arıza veya yeniden işleme oluşumları olan daha karmaşık sistemleri incelemek için iki fazlı Coxian işlem süreleri dikkate alınmıştır. Çalışmanın amacı, uzun vade ortalama sistem maliyetini en aza indiren eniyi kontrol politikasını bulmaktır. Markov karar süreci kullanılarak eniyi politikaların yapısı ortaya konmuş ve çalışma çeşitli sayısal örnekler ile zenginleştirilmiştir. İkinci olarak, eniyi kontrol politikaları bulmada karşılaşılan zorlukların üstesinden gelmek adına uygulaması kolay bir politika önerilmiştir. Önerilen politika birçok durumda eniyi politikaya yakın performans göstermektedir. Performansı eniyi politikadan uzak olan durumları iyileştirmek adına, önerilen yaklaşımın geliştirilmiş bir versiyonu da dikkate alınmıştır. Tez kapsamında yapılan son çalışma, iki-fazlı Coxian işlem süreleri, paralel makineler ve sonlu tamponlar içeren üretim hatlarının kesin bir Markov analizini sunar. Hammadde tedariği ve son ürün talebinin bağımsız Poisson süreçleri uyarınca geldiği bu problem, sürekli zamanlı bir Markov zinciri olarak modellenmiş ve geçiş hızı matrisini oluşturmak için özyinelemeli algoritmalar önerilmiştir. Genel özyinelemeli form 3-istasyon 4-tampon sistemlerine özgü olmasına rağmen, durum sayısını hesaplama ve durumları üretme rutinleri herhangi bir M-istasyon (M+1)-tampon sistemi için çalışmaktadır. Geliştirilen model, kararlı durum dağılımını ve verim, sistemdeki ürün sayısı ve ortalama sistem maliyeti gibi performans ölçütlerini hesaplamaya olanak sağlar. Önerilen metodoloji, daha uzun hatların performans analizi için bir ayrıştırma bloğu olarak da kullanılabilir.
Citation - WoS: 3
Citation - Scopus: 4
Control of M/Cox-2/s make-to-stock systems
(RAMAZAN YAMAN, 2020) Ozgun Yucel; Oender Bulut; Yücel, Özgün; Bulut, Önder
This study considers a make-to-stock production system with multiple identical parallel servers fixed production start-up costs and lost sales. Processing times are assumed to be two-phase Coxian random variables that allows us to model the systems having rework or remanufacturing operations. First the dynamic programming formulation is developed and the structure of the optimal production policy is characterized. Due to the highly dynamic nature of the optimal policy as a second contribution we propose an easy-to-apply production policy. The proposed policy makes use of the dynamic state information and controlled by only two parameters. We test the performance of the proposed policy at several instances and reveal that it is near optimal. We also assess the value of dynamic state information in general by comparing the proposed policy with the well-known static inventory position based policy.
Control of M/Cox-2/s make-to-stock systems
(2020) Özgün Yücel; Önder Bulut; Yücel, Özgün; Bulut, Önder
This study considers a make-to-stock production system with multiple identicalparallel servers fixed production start-up costs and lost sales. Processing times are assumed to be two-phase Coxian random variables that allows us to model thesystems having rework or remanufacturing operations. First the dynamicprogramming formulation is developed and the structure of the optimal productionpolicy is characterized. Due to the highly dynamic nature of the optimal policy asa second contribution we propose an easy-to-apply production policy. Theproposed policy makes use of the dynamic state information and controlled byonly two parameters. We test the performance of the proposed policy at severalinstances and reveal that it is near optimal. We also assess the value of dynamicstate information in general by comparing the proposed policy with the wellknownstatic inventory position based policy.
Citation - WoS: 1
Citation - Scopus: 6
Control of make-to-stock production systems with setup costs
(GAZI UNIV FAC ENGINEERING ARCHITECTURE, 2020) Sinem Ozkan; Onder Bulut; Özkan, Sinem; Bulut, Önder
This study considers the joint control of production and inventory in a Markovian make-to-stock environment with parallel production lines production start-up costs several customer classes and lost sales. At the production completion and demand arrival epochs there are two types of decisions. The first is the determination of the number of lines that should be active (producing) and the second one is whether to satisfy an arriving demand. Especially at the states where inventory position is low it would be more economical to reject the arriving demands of lower priority classes and reserve on-hand stock for future demands of more valuable classes. This study contributes to the control literature of make-to-stock systems with the simultaneous consideration of fixed production costs multiple lines and several customer classes. The dynamic programming formulation of the system which we model as a Markov Decision Process is developed by defining a two-dimensional system state vector. The model is solved with the value iteration algorithm. It is shown with the numerical examples that the optimal production and stock rationing policies have dynamic structures. Due to the dynamic nature of the optimal policies we also propose easy-to-apply and well-performing alternative policies. The proposed production policy is an extended version of the twocritical-number policy which is frequently considered in the literature. The proposed rationing policy is a dynamic one that considers the value of outstanding production orders in addition to inventory level.
Depo optimizasyonu ve organizasyonu: Örnek olay incelemesi çelik döküm şirketi
(2019) Çelik, Sezgi; Bulut, Önder; Kılıç, Gökhan
Pickiness in the consumers' behaviors forces the enterprises to produce 'customer oriented' products and services that meet global standards. In the developing global market, the importance of the customer orientation and satisfying the customers' demands within short time increased considerably. In the global market where the product qualities are very close to each other and they have nearly the same contents therefore timeframe has become the selective criterion in most cases. The purpose of this study is to organize the pattern warehouse of a steel casting company located in Izmir. Finding patterns, which are kind of molds, in the warehouse in a reasonable amount of time saves material, labor and time. Thus, the reduction in the time required to search for a pattern and transport it to the production area reduces production costs. While handling those problems the frequency of the orders by customers of the patterns and the physical conditions of the warehouse are considered. A well-organized warehouse layout and a quick transfer of the patterns are targeted throughout the study. Furthermore, in order to maintain well-functioning of the system an application of RFID technology is proposed. Key Words: Warehouse, Shelf allocation, optimization, mathematical model, RFID, Warehouse organization, layout change
Citation - WoS: 3
Citation - Scopus: 5
Exact analysis of production lines with Coxian-2-distributed processing times and parallel machines
(Elsevier Ltd, 2021) Özgün Yücel; Önder Bulut; Yücel, Özgün; Bulut, Önder
This study considers production lines where each station has parallel machines with 2-phase Coxian processing times. The setting in this paper is designed specifically for scrutiny of the replenishment of raw materials and finished goods inventories with intermediate buffers in-between stations. Each buffer has a limit of capacity. Raw material supply and demand for finished goods are generated according to independent stationary Poisson processes. Coxian-2 processing times can be utilized to model failure-prone machines with exponential service times times to failure and repair times. The second phase of Coxian-2 can also be considered as a rework operation visited with a predefined probability. We model the line as a continuous-time Markov chain and propose recursive algorithms to generate the transition rate matrix. Although the general recursive form is specific to 3-station 4-buffer lines routines for calculating the number of states and generating the states work for any M−station (M + 1)-buffer systems. The developed model allows obtaining steady-state distribution and performance metrics such as throughput the average number of items in the buffers and average system cost consisting of production holding and shortage costs. Furthermore we enrich our study with numerical experiments and analyze the impacts of buffer capacities processing rates of the machines and the number of parallel machines on the system performance. Moreover the exact analysis provided in this paper can also be used as the decomposition block for the performance analysis of longer lines. © 2021 Elsevier B.V. All rights reserved.
Gıda endüstrisinde stok yönetim sistemi
(2013) Akıncı, Efe; Bulut, Önder
Gıda Sanayinde müşteri tercih ve beklentilerinin değişimiyle birlikte taleplerdeki dalgalanmalar şirket kaynaklarının optimum kullanımı açısından önemli hale gelmiştir. Kaynak planlamalarının doğru ve tutarlı yapılabilmesi için taleplerdeki değişimler doğru tahminlenmeli ve stok yönetimi bu tahminler doğrultusunda yapılmalıdır. Bu çalışmayla birlikte gıda firmasının satış, stok ve plan tahminlerinin yönetimine ilişkin yeni bir ölçümleme metodu ile istatistiksel analizi hazırlanmıştır. Öncelikle malzeme grupları bazında geçmiş planlar ve stoklar incelenmiştir. Yönetim sürecini iyileştirmek adına geliştirilen yeni yöntemin yeterliliği istatistiksel analizlerle ölçülmüştür. Çalışma sonunda analizlere bağlı olarak yönetsel öneriler sunulmuştur. Firmanın ürün yönetim yöntemlerine uygun olarak geliştirilen yeni yöntem kullanılmaya başlanmıştır. Çalışmada yer alan firma diğer ürün grupları için de yeni yöntemi uygulamaya almayı planlamaktadır.
Gıda firmasında müşteri memnuniyeti ölçümlenmesi
(2012) Eker, Elvin; Bulut, Önder
Gıda sanayinde müşteri tercih ve beklentilerinin geniş çaplı değişimiyle birlikte müşteri memnuniyeti en önemli kriterlerden biri haline geldi. Müşteri odaklı yönetim mantığı ve müşteri memnuniyeti yönetim sistemi hızlı geri dönüşler sağlamaktadır. Müşteri memnuniyetinin düzenli olarak belli dönemlerde ölçülmesi ilgili sistemin gereğidir. Bu çalışmayla birlikte, gıda firmasına ilişkin satış ve lojistik aktivitelerin anket uygulamalı bir müşteri memnuniyeti ölçümleme metodu ile istatistiksel analizi hazırlanmıştır. Öncelikle farklı sınıflardaki müşteriler gruplandırmıştır ve müşteri beklentilerine ve ihtiyaçlarına göre özel ölçümler yapılmıştır. Çalışma sonunda analizlerimize bağlı olarak yönetsel öneriler sunulmuştur.Firmanın ürün ve hizmet servisleriyle ilgili geri bildirimlerinin müşteri memnuniyeti ölçümlenmesiyle sağlanabileceği bir müşteri geribildirim kanalı düzenlenmeye çalışılmıştır. Çalışmada yer alan firma müşteri ilişkileri yönetimine ilişkin düzenleyici önleyici faaliyetler almayı planlamaktadır.
Hotel overbooking capacity rationing and cooperation with third-parties: a two-period optimisation model
(Inderscience Publishers, 2024) Nazlı Karatas Aygün; Önder Bulut; Aygün, Nazli Karataş; Bulut, Önder
We propose a two-period optimisation model for a hotel revenue management (RM) problem where overbooking capacity rationing and cooperation with third-party websites are simultaneously considered. In a Stackelberg game structure the hotel first sets the price and overbooking and rationing levels and as the followers third-parties decide their effort levels by a Nash game. The proposed model is solved using a genetic algorithm. An extensive numerical study is performed to investigate the effects of multiple night stays hotel effort level and hotel capacity on the decisions and the hotel profit. It is shown that the value of capacity rationing increases with multiple night stays and the expected profit of the third-parties is decreasing with the hotel effort level but the relation between the hotel effort level and profit is not monotone. As the hotel capacity is expanded the effort level of the third-parties and the hotel profit increase. © 2024 Elsevier B.V. All rights reserved.
Kopmalara eğilimli akış ağlarında sistemik risk analizi ile bir kapasite artış planlama çerçevesi
(2020) Aygün, Nazlı Karataş; Bulut, Önder; Bıyık, Emrah
We propose a capacity expansion framework to guarantee a certain service level in failure-prone flow-networks composed of supply, demand and intermediate nodes, and arcs in between. We formulate the problem as a general stochastic optimization model to minimize the total cost of additional edge capacities. Our model allows considering different cost structures corresponding to the nature of different real-life applications. We consider a non-linear cost structure that captures both the immediate cost of investment and the cost of future risk. The feasible region is composed of the additional edge capacities that satisfy a probabilistic constraint (certain service level) which is the probability of total demand-not-satisfied is greater than a predetermined level is less than a risk threshold and a boundary constraint which is the additional edge capacities are greater than or equal to 0. These constraints are based on a systemic risk measure. Systemic risk measure, in contrast to traditional risk analysis on the component level, captures a holistic view of the system-wide operation and calculates the risk of unsatisfied demand. The requirement for system-wide analysis arises from the complex and nonlinear relationship between components of the system. Systemic risk is due to initial edge failures and the cascade of these failures throughout the system. Due to the stochastic nature of initial failures and certain supply and demand nodes, the distribution of unsatisfied demand is calculated via simulations embodied within a Grid Search Algorithm that identifies the feasible region. Cost-optimal edge capacity expansion is computed by a population-based heuristic optimization algorithm, namely, Differential Evolution. Our methodology can be applied to numerous fields including financial systems, power systems and supply chains. We apply our proposed framework to a medium-size general flow-network, and conduct a comprehensive numerical study to investigate the effects of system parameters on the feasible region and optimal solution.
M Coxian-2 s stoğa-üretim sistemlerinin kontrolü
(2016) Öztürk, Özgün; Bulut, Önder
In this thesis, we consider a make-to-stock production environment with multiple processing channels, several customer classes, fixed production start-up costs and lost sales. Demands of customer classes are generated from independent Poisson processes. Processing times are assumed to be independent two-phase Coxian random variables. Each phase of Coxian distribution is an exponential random variable corresponding to a specific stage in production and there is a certain visiting probability from phase-one to phase-two. Phase-type processing time assumption allows to model a system with a rework/inspection operation. The problem is to control the production and allocate the on hand inventory among different customer classes. We extend the production-inventory control literature by considering phase-type production times, several customer classes, parallel production channels and start-up cost in a single model. First, the dynamic programming formulation is developed and optimal production and rationing policies are characterized under average system cost criterion. Furthermore, a dynamic rationing policy and several production policies are proposed and their performance analyses are carried out. The final contribution of this thesis is to propose a new method, based on renewal theory, to calculate the long-run average system cost under the optimal production and static rationing policies when there is a single processing channel.