Reverse logistics of refillable glass bottles: a simulative approach PDF

Preview

Summary

Reverse logistics of refillable glass bottles: a simulative approach Antonio Cimino(a), Francesco Costantino(b), Giulio Di Gravio(c), Francesco Longo(d) (a) (d) Modeling & Simulation Center - Laboratory of Enterprise Solutions (MSC – LES) M&S Net Center at Department of Mechanical Engineerin...

Description

Accelerat ing t he world's research.

Reverse logistics of refillable glass bottles: a simulative approach francesco costantino Proceedings of the 2009 …

Cite this paper

Downloaded from Academia.edu 

Get the citation in MLA, APA, or Chicago styles

Related papers

Download a PDF Pack of t he best relat ed papers 

DET ERMINANT S OF INPUT, WORK-IN-PROCESS AND OUT PUT INVENT ORY LEVELS: AN INT ERN… Ruggero Golini

Reverse logistics of refillable glass bottles: a simulative approach Antonio Cimino(a), Francesco Costantino(b), Giulio Di Gravio(c), Francesco Longo(d) (a) (d)

Modeling & Simulation Center - Laboratory of Enterprise Solutions (MSC – LES) M&S Net Center at Department of Mechanical Engineering University of Calabria Via Pietro Bucci Rende, 87036, ITALY (b) (d) {a.cimino, f.longo}@unical.it (b) (c)

Operations Management Group Department of Mechanical and Aeronautical Engineering University of Rome “La Sapienza” Via Eudossiana 18 Rome, 00184, ITALY (b) (c) {francesco.costantino, giulio.digravio}@uniroma1.it

Keywords: Closed Loop Supply Chain, Reverse Logistics, Refillable Bottles, Traveling Salesman Problem Abstract This paper focuses on an integrated simulation model of a closed loop distribution system of water refillable bottles supply chain. In particular, the target of the research consists in evaluating the impact that the introduction of new nodes in the logistics configurations can have on service level to customers and on resources allocation, programming and scheduling. A case study of a distributor system that serves a limited geographical area is presented, to identify its performance parameters and how their variations can influence immobilizations, stock-outs and procurement time of restaurants served and fill production requirements. 1.

INTRODUCTION Christopher (1998), as others, defines supply chain as “the network of organisations that are involved, through upstream and downstream linkages, in the different processes and activities that produce value in the form of products and services in the hand of the ultimate consumer”. Design, modelling and analysis of supply chains is so focused on optimization of physical and informative flows, from supplying to distribution to customer.   Nowadays the concept of “environmental quality” is evolving its meaning from the original “cleaner environment” to “sustainable development”, where higher and wider attention is given to cover all the organization processes and, at the same time, improve the image of production and service systems. The purpose of extending traditional supply chain concepts is to allow assessment and management of

environmental impacts, present or potential, direct or indirect. Since many years, life cycle analysis has been a deeply studied research field, regarding the observation of environmental performances of products and processes from raw materials extraction to waste recovery (Lamming and Hampson, 1996). Many authors (Cattanach et al., 1995, Zhang et al., 1997) proved benefits of environmental improvements in costs, risks and responsibility reduction, safety and clean increasing, as well as social benefits. Greening the supply chain needs to face the issue of adding the traditional scheme some dedicated agents and elements, for collection, re-use, remanufacturing and recycling processes, to define a closed loop (figure 1). The performance of these processes depends on many factors, i.e. collecting capability, economic advantage of remanufacturing, presence of markets for recycled materials. To this extent, organizations need a framework to implement and evaluate actions of environmental improvements, able to grant success (Beamon, 1999).

Figure 1. Closed loop supply chain

Reverse logistics of refillable bottles is an always actual issue of environmental and sustainable development as reuse of glass containers is preferable to recycling according to the waste hierarchy: returnable container systems allows particular packaging to come back into the production cycle, avoiding to become wastes. Glass is an high quality material that preserves taste and organoleptic characteristics of beverages , the reuse of bottles (up to 50 times) grants huge savings of raw materials and energy and, when damaged or consumed, they can be 100% recycled with a composition of 60-80% in new bottles, even if this cost can reach five times the one of cleaning and sterilization. Nowadays, more than 60% of drinks are in glass where the consumer has to pay only a fee of 15 to 25 cents for each bottle that is given back when the bottle is returned (generally a further caution for plastic cases is needed). To this extent, opportune strategies of reverse logistics in closed loop supply chains are to be configured, depending on the different level of centralization and decentralization of the agents and on the possibility of defining pooling groups of customers, wholesalers and distributors (Blumberg, 2004). This generates a complex problem of resource allocation (Stock et al. 1998) that involves transportation scheduling, optimization of routes (Dethloff, 2001), less than truck-load management, warehouses capacity (Minner, 2001) and customer satisfaction (Fleischmann et al. 2000 and 2001). Decision support systems has to be implemented to help managers in network design process, integrating optimization techniques and simulation models to identify critical success factors of any reconfiguration activity. 2.

PROBLEM DEFINITION The logistics system defines the direct and reverse streams of refillable glass bottles, in a three echelon closed loop supply chain. The case study represents a real scenario acting in the south of Italy territory, in the province of Latina studied in depth to collect source data for the simulation model. The actual configuration (figure 2) is made of a single distributor (the red pin) serving 10 restaurants (blue marks) with very different demands of bottles of water, from an average of 3 to 18 cases per day with a standard deviation of 25%. Every day the distributor defines its best trip to satisfy at most orders received, considering only one vehicle with a maximum capacity of 200 cases (6 bottles per plastic case), collecting and taking back empty bottles that have to be returned to the supplier. A 1 to 1 kanban-like logic is implemented (a case of empty bottles retired for each one of full), considering the 2% of broken or damaged bottles not affecting the system as the cases are anyway to be given back.

Figure 2. Problem context From this starting situation the distributor has the opportunity to add other 5 restaurants (yellow marks) depending on the impacts that they would have on its performance parameters, considering also reverse streams. The new customers are partially distributed on consolidated routes but also cover different areas. The distance matrix is defined considering different speed of motorways and local routes where these latter are only limited to very short paths of the same town. 3.

THE DECISION SUPPORT SYSTEM The problem needs a decision support tool to identify, with the resources given, supply chain capability and flexibility levels. The model presented consists of two different stages. The first step runs a simulation model of the logistics network with stochastic demand according to direct data collection; the output of the different trials allow a recursive identification of needs in terms of number, frequency of orders and inventory of full and empty bottles for both restaurants and distributor’s storage. The second step starts from the scheduling of the orders to solve a Travelling Salesman Problem with a minimum consumption and loading saturation principle. 3.1. The simulation model The simulation model, implemented with the SIMUL8 suite, presents for each restaurant element: • two storage bins for full and empty cases; • a customer demand distribution probability, coming from on field interviews and data collection; • a standard reorder policy of continuous review stock, where the Reorder Quantity (Q) is calculated in a deterministic way (the quantity necessary to bring the inventory up to a certain recovery level), according to the defined Reorder Point (R). Every restaurant is so a parametric entity where its reorder strategy depends on the parameters in Table 1. Data gathered from simulation are obtained with an iterative and recursive process, assigning a starting value to parameters and tuning them according to the

results of each trial (for example, the average order fulfillment lead time L passed from a starting value of 1 to 2 days with a standard deviation of 1). The recovery level is strictly related to the Reorder Point: according to the order fulfillment lead time and considering the inopportunity of issuing multiple or consecutive orders, the Reorder Quantity needs to cover a correct time extension. The k value of Safety Stock is high because stock outs of water in restaurants has to be constantly avoided: eventual emergencies are generally solved with instantaneous direct orders to the nearest possible supplier and are not considered in this model. Table 1. Restaurant characterization Parameter Formula or source on field Demand (D, σD) Order fulfillment lead time from simulation (L, σL) Reorder Point (RP)

RP = D·L + SS

Safety Stock (SS)

SS = k D2 ⋅ σ L 2 + L ⋅ σ D 2

Recovery Level (RL) Reorder quantity (Q)

k=2 (97,7% service level) RL = 2,5 RP Q = RL - GS

Goods in Stock (GS)

dynamic, from simulation

Service level factor (k)

The distributor element has to receive and fill customer orders, collect empty bottles and give them back to its supplier. The main elements are: • an order scheduler; • a vehicle that serves the whole area; • a storage bin for returned empty cases. During its activity the distributor has to consider as performance parameters: • service level offered to restaurants, in terms of order fulfillment time and stock out protection; • inventory level of returned bottles: according to the agreement with its supplier, the storage is emptied twice a week; • costs of shipment, trying to saturate every load and drive down fuel consumption. Timing in the simulation follows steps in figure 3. Orders typically arrive in the morning, during a predefined time window, so to organize the shipment for the next day. Every morning the vehicle is loaded with quantity and travel sequences defined the day before; orders that arrive before the loading activity are included while the ones after that moment automatically shift to the next day. Due to this organization and distances covered, the minimum time for order dispatching is 1 day with a single trip arranged every day.

Figure 3. Model timing 3.2. Order scheduling and TSP solver The scheduling has to be defined trying to fill up to the load capacity so to increase vector saturation and reduce the number of trips to completely fulfill the orders received. Anyway, it’s also necessary to avoid stock outs of restaurants, so the truck consolidation process has to be balanced with a prioritization principle. The scheduler algorithm takes orders with a FIFO logic, progressively adding cases to vector, then, if an order cannot be added due to capacity limits (dividing orders can be considered inopportune as it would increase the number of travels), it is moved to a backlog storage bin and it will be the first loaded in the following shipment. Once the list of restaurants to reach is given, the consignee sequence could be solved with the typical Travelling Salesman Problem approach, where the task is to find the shortest possible path to visit every destination, leaving from and coming back to a base point. Considering that restaurants are not so far from each other, it’s to find a solution that could grant economic advantages reducing at most the fuel consumption, that means trying to empty vehicle as soon as possible. According to this strategy a greedy heuristic is implemented, where, starting from a knot of reference, the optimal sequence is constructed simply passing on the nearest knot. For every trip the solver algorithm (implemented in Visual Basic) starts from distributor and looks for the closest destination from a distance table, considering only ordering restaurants. 4.

SIMULATION RESULTS The starting scenario considers the actual situation with the distributor serving 10 restaurants and collecting network performance levels. After the validation, the study introduces further 5 restaurants to stress the system and extract supply chain flexibility. The two

experiments are tested on 20 trials of summer most demanding 100 days of the year. The first set of results looks at distributor’s service level in terms of order fulfillment lead time (table 2). We can notice that restaurants are served with a small variation in average waiting time and less variance in the process, as the increased number of customers generates more opportunities to fill up the vehicle and serve them within a journey: simply, it’s easier to find small orders to complete the truck capacity. Number of stock outs has to be checked as it’s related to the definition of the reorder policy parameters (such as the k value) and during the simulation model a trial & error test was necessary to tune it until all restaurants had a generally continuous service. These parameters can so be considered invariant as the service level it’s constant once defined that every process has a dedicated time window in the model. Table 2. Order fulfillment lead time 10 rest. 15 rest. variation L (hours)

43,0

46,5

+8%

σL (hours)

23,1

20,6

-11%

# of stock outs

~0

~0

Second set of results looks at covered distances to guarantee this service level (table 3). Of course, the total covered distances increases in a significant way, but coherently with the growth of restaurants (44% vs. 50%) and there is a sort of economies of scale with an 8% of cost reduction. It’s to notice that the solution doesn’t increase much the average distance of a single journey (+13%) but tends to increase more the number of journeys (+28%), also due to a better alignment of orders in time. The minimum covered distance shows in both scenarios there are still one-to-one consignee and increasing the number of customers do not repair this efficiency problem, where vehicle travels just to avoid one restaurant stock out. As for reverse stream load logics (explained later) dedicated solutions could be implemented to manage these exceptions Table 4 shows information about the vehicle utilization. The increase of the average number of cases transported is remarkable and could strongly suggest to implement an aggressive policy so to get new customers, maybe even more than the five presented. Nevertheless, the reverse stream of empty bottles needs to consider that, starting a shipment with more than 75% of capacity loaded, generates difficulties in the handling process of load/unload. As every stop the carrier loads empty bottles, each time their amount surpasses the number of full ones there’s a risk of blockage of a FIFO one-entry-point system.

In this second situation, the percentage of trips with more than 75% of loaded capacity nearly doubled. Considering reverse stream of bottles, the increase of customers is still possible without new resources and brings some advantages of saturation but creates general inefficiencies that could affect delivery times. Table 3. Covered distances 10 rest. 15 rest. covered km (total) covered km per journey (mean) covered km per journey (st.dev.)

variation

5395,0

7759,8

+44%

88,0

99,0

+13%

41,7

41,5

-1%

min covered km

13

27

max covered km

202

224

# of journeys

61,4

78,4

Table 4. Vehicle filling 10 rest. 15 rest. Vehicle filling (mean) % of full truckload Vehicle filling (st.dev.) % of full truckload % journey with more than 75% of utilization

104,7

159,8

52,3%

79,9%

63,4

31,5

31,7%

15,7%

36,5%

69,4%

+28%

variation +53%

-50%

+90.3%

The reverse stream of empty bottles needs distributor to set up a large storage area with appropriate handling system for those fragile goods with associated warehouse managing costs. It is to recall that every water supplier defines accurate agreements with distributors on empty bottles return policies, in terms of quantities, frequency and their variability, due to huge savings of raw materials and energy that can be realized. The simulation model monitors the level of empty bottles inventory as its contents is directly proportional to storage cost. Figure 4 shows the trend of this quantity in dependence of time, illustrating the effect of a standard covenant of total recollection in fixed frequency, that is the actual situation of the area where a single trailer can anyway load the total amount in the two presented situations. A similar tendency (figure 5) can be identified in each restaurant where the recovering frequency is not determined, depending on

the different orders issued, and the inventory isn’t regularly emptied.

turnaround time (both for full and empty bottles), and the general increase of the order fulfillment time suggests new models of order scheduling to improve distributor performances. Table 5. Reverse stream immobilization (1/3) Distributor storage area (empty bottles) -95%

Average

+95%

10 restaurants scenario Average queue size (cases) Maximum queue size (cases)

495,55

521,23

546,92

1114,40

1163,90

1213,40

15 restaurants scenario

Figure 4. Distributor reverse stream inventory

Average queue size (cases) Maximum queue size (cases)

809,61

826,46

843,30

1749,14

1782,70

1816,26

Table 6. Reverse stream immobilization (2/3) Low demand restaurant storage area (empty bottles) -95%

Average

+95%

10 restaurants scenario Average queue size (cases) Maximum queue size (cases) Average queuing time (hours) St. Dev. of queuing time (hours)

22,39

22,77

23,14

35,24

35,80

36,36

179,45

184,10

188,76

58,91

59,82

60,72

15 restaurants scenario Figure 5. Restaurant reverse stream inventory Averages value passed from about 521,23 to 826,46 cases, with a total increase of 59%, and a further increase of the maximum queue size of 53%, from 1163,9 to 1782,7 cases (table 5). These two parameters are necessary to consider interventions in capacity of the storehouse and infrastructures or to reconsider agreements with suppliers. The same effect can be noticed on restaurants as a development of the set of served could have the same consequences on their empty bottles inventory. Table 6 and 7 illustrate the evolutions of the immobilizations in the network, before and after the supply chain extension. Queue sizes remains at similar levels for small restaurants and increase of 6% for bigger ones with minimum values that tends to decrease. Queuing time, generally lower for bigger orders, tends to decrease for low demand restaurants (-3%) and increase for high demand ones (+6%), reducing the spread of the two typologies of customers (-38%). As expected, the saturation of the vehicle capa...