Collaboration between freight carriers has become an important mean to remain competitive in the less than truckload (LTL) industry. External pressures such as worldwide competition and fuel prices have driven carriers to establish coalitions to balance the transportation resources and reduce operational costs. This thesis proposes a mechanism for small and medium-sized carriers to support the collaborative planning decisions. The aim of the proposed mechanism is to minimize transportation costs by enabling specialization of customer areas through the introduction of transshipments between the facilities of collaborating carriers. The approach is tested by means of a simulation study using real case data. Results show that LTL carrier collaborations can realize considerable reductions in the transportation costs.
Intra-Enterprise Collaborative Planning
With Transshipments
Master’s Thesis
1st of August, 2013
Author: Jose Alejandro Lopez
Supervisor: Prof. Dr. K. J. Roodbergen
Table of Contents
1. Introduction: ... 2
2. Literature Review: ... 3
3. Research Objectives: ... 6
4. Methodology: ... 6
4.1 Research method outline: ... 6
4.2 Justification: ... 7
4.3 Definition of Performance measures: ... 7
4.4 Analytical model development: ... 8
5. Modeling Formulation: ... 8 5.1 Problem definition: ... 8 5.2 Objective function: ... 9 5.3 Parameters: ... 9 5.4 Variables: ... 9 5.5 Constraints: ... 9 5.6 Assumptions: ... 10 6. Exchange Mechanism: ... 10 6.1 Overview: ... 10 6.2 Pseudo code: ... 11
6.3 Explanation of the mechanism ... 11
6.4 Considerations of the mechanism: ... 13
7. Simulation study:... 13
7.1 Simulation setup: ... 13
7.2 Practical Case: ... 14
7.3 Simulation Results: ... 15
7.3.1 Quality of solutions: ... 15
7.3.2 Computational complexity performance: ... 17
7.3.3 Robustness:... 17
7.4 Practical implications... 19
7.5 Theoretical implications ... 20
8. Conclusions and further research. ... 20
1. Introduction:
Worldwide competition of less than truckloads (LTL) carriers represents a threat for the small and medium enterprises (SMEs) that are competing in this industry. Not only larger logistics networks and broader services, but also economies of scale and scope are some of the advantages of the large companies within the sector (Bloos and Kopfer, 2008). On top of that, internal responses to such external pressure have already been intensively optimized in most organizations (Skjoett-Larsen, 2000). As a result, external collaboration for small and medium-sized carriers has become a necessity in order to remain competitive (Krajewska and Kopfer, 2006). In general, two different perspectives are recognized in external relationships of freight carriers: vertical cooperation among carriers and shippers, and horizontal cooperation among multiple carriers. In the light of the latter, carriers can efficiently exchange customer orders (transportation orders for goods to be collected and delivered at specific points) with the purpose of generating cost savings that cannot be achieved acting individually (Kopfer and Wang, 2011). Ergun et al. (2007) suggest that a collaborative relationship between freight carriers can increase the load factor of trailers and reduce the empty hauling. For instance, collaborative freight carriers’ partnerships can achieve a reduction in transportation costs between 5 and 15% (or more) through order sharing and joint planning (Cruijssen and Salomon, 2004).
Interestingly, although the conclusions about the economic benefits of horizontal cooperation among freight carriers are widely accepted, research on the implementation of this type of collaboration is on an early stage. A survey among logistics service providers (Cruijssen et al., 2007) identified that one of the challenging issues in the practical context of horizontal cooperation is to define the allocation of customer orders among the members of the collaborative group (i.e., determining a workload allocation plan). The main aim of this allocation plan is to minimize the total fulfillment costs of the partnership by making an efficient use of the transportation resources of the group as a whole. This objective, however, underlies a very complex problem in which several carriers’ operations should be considered simultaneously. As a consequence, in practice, this allocation plan is often the result of unilateral decisions of partner carriers to subcontract cost-intensive customer orders, rather than a collaborative planning in which decisions are made jointly to minimize the total operational costs. Consequently, collaborative groups, in general, perceive lower economic benefits than their potential.
that are based on electronic transactions between carriers. As a result, the physical transportation routes through which carriers physically exchange their goods are either disregarded as a source of transportation costs and coordination (e.g., Cruijssen and Salomon, 2004), or omitted as a form of collaboration (e.g., Krajewska et. al, 2008) in which carrier partnerships can reduce operational costs by complementing efficiently their operations (Hernandez and Peeta, 2010).
Taken the abovementioned considerations in mind, this research project intends to develop a simple, implementable and economically effective workload allocation mechanism for collaborative groups of small and medium–sized carriers in the LTL industry. The approach will consider a LTL carrier network of two business divisions of a holding company that operate in overlapping customer areas. In contrast to what has been previously proposed in literature, this research will consider the physical flows among carriers by introducing line haul shipments (i.e., direct transportation routes between their facilities). The main rational of the approach that will be proposed is to achieve collaborative benefits through the specialization of customer areas in order to avoid overlaps in the transportation operations. The performance of this mechanism, which will reflect the impact of transshipments on the transportation costs of the group, will be determined by means of a simulation study using real case data.
The remainder of this thesis is organized as follows: section 2 will provide a literature review of the different approaches available to allocate workload in freight carriers’ collaborations. In sections 3 and 4 the research objectives and the methodology chosen to meet our research objectives are explained. Section 5 will proposes the problem definition of the collaborative context that is considered in this research. Thereafter, section 6 will describe the solution designed to allocate workload. Section 7 will present the simulation studies and, finally, section 9 will present the conclusions and will propose areas further research.
2. Literature Review:
Figure 1 –Reduction of total vehicles and miles [adopted from Kopfer and Wang, (2011)] In addition, economies of scope are obtained by decreasing the number of miles traveled when a variety of routes, with overlapping or adjacent customer areas, are rearranged efficiently. Figure 2 represents this scenario.
decided not to include within their own fulfillment plan. Despite the fact that only a part of the operation is planned jointly, which diminishes the potential benefits of the group (Berger and Bierwirth, 2010), these decentralized approaches are needed in collaborative groups where carriers are not willing to disclose private information and lose planning autonomy (Kopfer and Prankratz, 1999; Krajewska and Kopfer 2006; Wang and Kopfer, 2010).
Krajewska and Kopfer (2006) suggested a framework for the design of decentralized mechanisms for the collaborative planning. The three phases of this framework are characterized as follows: a pre-processing phase, where each carrier decides which order requests are going to be offered to collaborating partners and their respective self-fulfillment cost; an optimization phase, where the pool of requests offered to the collaboration is allocated among the members of the group; and finally, a profit sharing phase, were economic benefits achieved by the coalition are divided. Most research in the decentralized stream of literature considered the allocation of customer orders, which refers to the optimization phase in the aforementioned framework, with multi-agent auctions models. Gomber et al. (1997), for example, proposed the Vickerey auction to allocate the different requests that each carrier offered to the coalition. In this auction, every carrier quotes a bidding price for each request without knowing the bidding quotes of other participants. As a result, bidders are incentivized to reveal their truthful valuation of the requests instead of quoting bidding prices that are high enough to win the auctions. This approach was further extended by Schonberger (2005) to allow participants of the auction to bid for bundles of requests instead of making quotes for individual orders.
In a further research, Schwind et al. (2009) presented workload allocation mechanism for the context of a holding company with business divisions that have overlapping transportation areas. Interestingly, the authors of that research decided to design a decentralized rather than a centralized approach due to the computational complexity of the resulting planning problem. The main rational of their proposed mechanism was to perform combinatorial auctions preceded by a predefined process based on a convex hull approach. The aim of the predefined process is to narrow down the number of requests to which each carrier can bid on the auction to minimize the complexity of the problem. The results in Schwind et al. (2009) show that transportation cost savings for the whole system by collaborating were approximately 14%. In another article, Berger and Bierwirth (2010) developed an iterative approach to efficiently allocate single requests (or bundles of requests) through a feedback process nurtured by the marginal profits reported by each participant in every iteration. The computational results show that the greater the overlapping area between carriers, the higher the potential benefits the group can obtain.
plan disregarding the physical transshipments between carriers. While in the other hand, decentralized approaches are market based allocation mechanisms grounded on electronic transactions. As a Consequence, the economic impact of physical transshipments in LTL collaborative groups has not been considered in the existent mechanisms to allocate workload.
3. Research Objectives:
The primary aim of this research is to develop workload allocation mechanism for LTL freight carriers’ collaboration where business divisions of a holding company exchange goods through line haul shipments. The line haul shipments to be considered on the mechanism will serve as a form of physical transshipments among carriers. Consequently, the secondary aim for this paper is provide insights when one includes these physical transshipments in collaborative planning problems.
In order to reach this objective, specific sub objectives must be achieved:
The definition of a criterion that allows classifying each customer order on the basis of how appealing they are for a reassignment based on specialization of customer areas. This criterion will serve to define the customer orders that will be loaded on the line haul shipments.
Define a feedback process in which the reassignment decisions are evaluated and adjusted according to the estimation of the total transportation costs that stem from both the local distribution routes and the line haul connecting the facilities of the carriers.
Estimate the performance of the approach in order to measure the potential value of introducing transshipments as a form of collaboration.
4. Methodology: 4.1 Research method outline:
4.2 Justification:
Transportation planning decisions, which constitute the underlying problem of the workload allocation mechanisms, represent a very complex task for academics and practitioners. On the one hand, attempts to optimally solve operational problems that are not efficient neither effective since these problems are NP-hard (Wang and Kopfer, 2010). On the other hand, simplistic rules of thumb may significantly diminish the potential economic benefits of collaboration. Consequently, research on workload allocation mechanisms has been addressing the problem by designing approximate algorithms in order to identify acceptable solutions in reasonable periods of time (e.g., Schwind et al., 2009; Kopfer and Wang, 2011). Moreover, with the aim of testing the performance of these mechanisms, simulation studies represent the standardized approach (e.g., Krajewska and Kopfer, 2006; Berger and Bierwirth, 2010; Kopfer and Wang. in progress) since this technique represents an excellent alternative to test the models before they are implemented in practical contexts at stake of economic resources. Accordingly, we adopt a research method that, in general, is consistent with prior research.
4.3 Definition of Performance measures:
The three principal performance measures of a heuristic approach are: quality of solutions, computational effort, and robustness (Barr et. al, 1995). Accordingly, each of these factors will be measured as follows:
Quality of the solutions: Previous research on workload allocation planning has measured performance by establishing upper and lower bounds to the total operational costs. The upper bound (i.e., the worst possible performance) comes from simulating the isolated planning scenario – where there is no collaboration - while the lower bound (i.e., the best possible performance) is defined by a complete joint planning (e.g., Kopfer and Wang, in progress; Berger and Bierwirth, 2010). Consequently, the quality of the solution will be measured with respect the upper bound as follows:
;
.
.
Computational effort: The computational time required to find the solutions by the mechanism will be reported and analyzed.
Robustness: The real case data that will be used in the simulation study contain several different days that vary in total number of customer orders, total volume of goods (specified in loading meters), and the average size per order. Therefore, the quality of solutions against these three factors will be analyzed.
Finally, in relation to the measurement of cost, this research will use the total kilometers travelled (including local delivery routes from both collaborating carriers and the line haul transportation routes to physically exchange goods) to fulfill the operation as a proxy. This approximation is justified since it reflects the most significant expenses incurred by LTL carriers to perform the operation: the fuel consumption, the drivers’ hours and the wear of tires. On top of that, a workload allocation plan has little impact on other type of LTL carriers’ operational costs.
4.4 Analytical model development:
In order to develop the heuristic-based analytical model this research will make use of the framework developed by Krajewska and Kopfer (2006). This framework, which is also used in e.g., Schwind et al. (2009) and in Wang and Kopfer (2011), not only suggests including an exchange procedure, but also considers a pre-selection process (preprocessing) in which carriers define pre-selection rules to define the most interesting orders to be reassigned to a carrier partner. The purpose of approaching the problem through this framework is to reduce the complexity of the problem by dividing it in simpler interrelated problems that interact with each other as in Schwind et al. (2009). The specifics of the approach designed will be shown on section 6.
5. Modeling Formulation:
This section describes the LTL collaboration problem for the allocation of workload between two business divisions of a holding company – assuming each division has an autonomous planning department that are willing to share full information.
5.1 Problem definition:
This research project is positioned in a logistic network of two business divisions of an LTL carrier holding company that are interested in cooperation. Each division has its own fleet of vehicles with homogenous capacity. Additionally, both carriers have their own customer order portfolio that must be fulfilled, for which transportation routes are planned by their autonomous planning department. The customer orders consist of goods that must be collected and delivered at specific points. After collections take place, the freight is transported to the carrier’s facility where the orders are processed and handled to be delivered on the next day. In the isolated planning scenario, each carrier defines a fulfillment plan for the distribution routes incurring into transportation costs of C1 and C2 respectively. In
transportation costs in the isolated planning (C1 + C2). It is assumed that the definition of the
collaborative plan is performed by a neutral planning unit that has access the complete information of both carriers. This assumption is justified by the existence of a holding company that relates both carriers in terms of ownership. Furthermore, the collaborative scenario considers physical flows between carriers through the definition of daily shipments that connect the facilities of both carriers (line hauls). Figure 3 shows an example of the operation on the isolated and collaborative scenarios.
Figure 3 – Introduction of line haul as a form of cooperation. 5.2 Objective function:
The main objective of the problem is the minimization of the total transportation costs of the complete operation of the collaborative group (i.e. Distribution routes costs and line haul shipments costs)
5.3 Parameters:
Complete requests portfolios with the specifications of the orders (volume, and location of collection and delivery points).
Geographical location of the carriers’ facilities. 5.4 Variables:
Division of the delivery orders between the business divisions. Total number of line haul shipments used (round trips). Total number of delivery trucks used.
Content of the line haul shipments 5.5 Constraints:
Number of delivery trucks available for each company. Number of line haul trucks available.
5.6 Assumptions:
All customer orders are eligible for exchange with the partner of the cooperation. In real
environments it is likely that companies have preference to serve some of their most important customers. However, introducing to the problem constraints related to individual interests increase the overall complexity of the workload allocation and may reduce the potential benefits of the group (Wang and Kopfer, 2010).
The total line haul trips are done before the trucks fulfilling the delivery routes start
their activities. This assumption implies that there are no precedence restrictions between the line haul trips and the delivery routes of the same day, which significantly reduces the coordination and planning complexity.
All the carriers share their complete and truthful information. This assumption implies
that carriers have aligned interests since they belong to the same holding company. Therefore, carriers will not attempt to jeopardize the collaborative welfare if it might go against their individual interests. The relevance of this assumption is based on the fact that asymmetric information can lead to solutions that cannot be guaranteed to be neither efficient nor accurate.
Collections are not operationalized. This assumption is based on the fact that the
central interest of our research is the definition of line haul shipments as a form of cooperation, and, in that sense, the collections done on one operational day will not influence the line haul shipments of the same day (if considered, collections would influence the line haul of the next day). This assumption, however, has an impact on the quantitative results of the total costs since they are going to be underestimated.
Time windows are not a restriction. In practical cases we see that LTL carriers are
subject to time restrictions based on a frame of days to fulfill the operation rather than specific time windows for a day, since the distribution routes operate during the standard working schedules of the week days.
The physical location of the customer orders prior to the workload allocation decision is
the facility of the carrier that received the order from the customer. This assumptions seeks to define a physical location of the order with the aim of consider the transfer operation by means of line shipments.
6. Exchange Mechanism: 6.1 Overview:
Input: CompleteOrdersList, AgregateAreas, Carriers Output: AssignationOfCustomerOrders, EstimatedTotalCost
1 for each AgregateAreas K do
2 Calculate the score of aggregate area K for carriers(1): =Score(carriers(1) ,K) = Distance(carriers(1), K) - Distance(carriers(2), K); 3 Calculate the score of aggregate area K for carriers(2):= Score(carriers(2) ,K) = Distance(carriers(2), K) - Distance(carriers(1), K); 4 next
5 for each Carriers n do
6 Create a ranking list of the aggregate areas for carrier i by ordering them downwards using score(n,aggregate area) as criteria; 7 next
8 LineHaulShipments = 0; 9 TotalCostActualIteration = 0 ; 10 TotalCostPreviousIteration = ∞; 11 Condition = false;
12 while ¬ condition = false do 13 for each carriers n do
14 Fill in the linehaul shipments of carriers(n) based on the ranking list of carriers(n) as criteria; 15 next
16 Create AssignationOfCustomerOrders list; 17 for each carriers n do
18 Solve the vehicule routing problem for carrier(n) based on AssignationOfCustomerOrders; 19 next
20 Calculate TotalCostsActualIteration: Linehaul Costs + Delivery routes costs; 21 If TotalCostPreviousIteration > TotalCostActualIteration do 22 Condition = true; 23 else 24 LineHaulShipments = LineHaulShipments + 1; 25 TotalCostPreviousIteration = TotalCostActualIteration; 26 end 27 end 28 FindTheCapacityUtilizationOfLastTruck()
31 return AssignationOfCustomerOrders, TotalCostActualIteration;
Pseudocode for the Wrokload Allocation Mechanism
6.2 Pseudo code:
Figure 4 – Pseudo code of the proposed mechanism. 6.3 Explanation of the mechanism
As mentioned in section 4.4 the proposed mechanism is based on the framework of Krajewska and Kopfer (2006) in the sense that it is composed of a pre-selection phase, in which customer orders are chosen to be part of the collaboration; followed by an exchange optimization phase, where the number of line haul shipments are determined. In order to make a detailed explanation, the complete algorithm is going to be explained in 5 steps. To further illustrate the steps of the mechanisms consider the example depicted in Figure 5. In this example two business divisions of a holding company (characterized as triangles) have to fulfill a set of deliveries on a given day of operations. The triangles represent the facilities of the carriers and the points the locations where the orders must be delivered.
Figure 5 – Logistic network of two carriers.
orders of both carriers. The purpose of this classification is to generate bundles of requests based on the geographical closeness of the delivery points, which enables the possibility of an efficient consolidation of freight for the routes. The greater the aggregate areas defined, the higher the tolerance to consolidate orders that are farther from each other. Figure 6 shows these geographical classifications of our example.
Step 2 (Define the savings criteria for each aggregated areas in relation to each company; lines 1-4 pseudo code): This step defines a distance based score that will serve as criterion to identify the most appealing aggregate areas to exchange to the partner carrier. The score, which is computed for each carrier independently, considers the distance between the aggregated areas to both carriers’ facilities. As a result, the score will serve as a proxy of collaborative welfare rather than a measure of unilateral utility.
Step 3 (Define the ranking list for each of the carriers; lines 5 – 7 pseudo code): Based on the scores computed in the previous step, a ranking list for each of the carriers is developed by prioritizing the individual customer orders, based on the aggregated areas to which they belong. Customer orders are prioritized from high (most interesting to exchange to the partner carrier) to low (least interesting to exchange to the partner carrier).
Step 4 (Iterative procedure of the defining line haul shipments and simulation; lines 12-27 pseudo code): In this step, a number of line haul shipments are defined (starting from 0 and then iteratively increased). Next, those line haul shipments are filled with customer orders based on the ranking list developed in step 3. Thereafter, a vehicle routing problem is solved for each of the carriers. At the end of this process, the total costs are evaluated and compared with the costs of previous iterations. If the costs are lower, the iterative process is repeated considering 1 more line haul shipment. Otherwise, if the costs of the iteration are higher than the cost of the previous, then the iterative process is finished. The goal of this procedure is to approach the local optimal by a neighborhood search (the neighborhood associated with the number of line hauls).
Figure 7 – Final assignation of customer orders 6.4 Considerations of the mechanism:
Four considerations are important to remark about the mechanism developed. The first one is that the iterative process searches for the minimum transportation costs assuming that the objective function is convex. The second consideration is that the fixed nature of the ranking classification locks the search procedure to remain permanently in the same neighborhood of the solution space, which means that the solutions obtained by the mechanism are suboptimal and highly dependent on the ranking list. Thirdly, it is necessary to highlight that not all the aggregate areas are completely assigned to a carrier; therefore, the resulting workload allocation defined by the mechanism will probably have some aggregate areas with overlap in the operations of both companies. Finally, the ranking list is randomly organized for the customers that belong to the same aggregate area.
7. Simulation study: 7.1 Simulation setup:
In order to rigorously measure the performance of the proposed workload allocation mechanism, a simulation setup must be explicitly specified. For the purpose of this research, a set of 200 days of data (taken from a real case scenario that will be introduced in section 7.2) were selected carefully in order to assure that the mechanism is tested through a wide range of different input parameters (i.e., total number of loading meters, total number of orders, geographical distribution of delivery locations). Nonetheless, due to resource limitation, the input related to the aggregated customer areas in which the customer map is going to be divided will remain fixed during the complete simulation study. Moreover, some of the input parameters do not change throughout the simulation due to the characteristics of the data (e.g., market power of each carrier and location of the facilities). In addition, the restrictions related to the fleet size and fleet capacity will be attained to the specifications of the case study (delivery trucks have a capacity of 13.6 loading meters, and line haul trucks of 24 loading meters).
mechanism; (c) carriers collaborate by means of commonly used exchange policies. Scenarios (a) and (b) are required to measure the quality performance of the mechanism, while the scenario (c) will serve as an experiment to test the economic advantages of supporting collaboration decisions by means of workload allocation mechanisms against other policies. The policies that will be analyzed in scenario (c) relate to fixed exchange agreements in which a pre-determined number of line haul routes taking place every day. The trailers operating these routes will be filled at a 100% of their capacity based on the same ranking list that was developed for the workload allocation mechanisms. These specific policies will consider the scenarios where carriers fix the number of line haul shipments to be 2, 3 and 4.
Finally, the simulation setup for the three scenarios mentioned before is developed in a Visual Basic environment according to a set of procedures that were carefully chosen for the objectives of this research. Specifically, we decided to approach underlying VRP by selecting the Nearest Neighbor heuristic. The selection criterion is the low computational complexity of the approach. The reason why this aspect is decisive is the fact that the mechanism proposed is designed to perform iterative procedures that involve several VRP problems. On top of that, the data selected consist of operational days that range from 300 to 1100 customer orders. As a consequence, any effective VRP heuristic approach would not be capable to find solutions in acceptable periods of time. Notwithstanding, this research acknowledges that the quality of the solutions is compromised due to this decision.
7.2 Practical Case:
This research is inspired on a case study of a holding company that seeks to establish collaboration between two small-medium size business divisions that operate in the LTL industry in the Netherlands. Each of the divisions has individual customer orders that consist of a set of collection and delivery points. A study of the data showed that around 95% of the transportation orders are collected, unloaded in the cross dock facility of the business division, and delivered the next day. The other 5% of the orders have a collection and a delivery point conveniently located to be served with direct transportation. Every day the trucks leave the facility in the morning loaded with orders to be delivered, and arrive in the evening with the freight collected on that same day. In general, most of the collection points are located around the facility of the carrier that receives the order; whereas the delivery points are spread across the Netherlands. An image of the operation (collection and delivery points to be visited) in one representative day for each of the companies can be seen in figures 8 and 9.
As it is possible to see, both companies operate in all the regions of the Netherlands. This situation represents an ideal scenario for this research since the overlap in customer areas open the opportunities to reduce transportation costs through collaboration. In addition, considering the fact that every day there are significant flows between the north and south of the country, we have a scenario where extra benefits can be obtained through use of line haul transshipments.
7.3 Simulation Results: 7.3.1 Quality of solutions:
The results of the simulation yielded positive outcomes in terms of the potential benefits of collaboration by means of transshipments (in this case as a form of line haul shipments).Table 1 presents the average performance measure in relation to the quality of the solutions (percentage of operational costs saved) for each scenarios. Scenario (a), which refers to the situation in which planning decisions are done in an isolated fashion, was outperformed by all the collaborative strategies considered. Scenarios (c) 2, (c) 3 and (c) 4 characterize the situation where companies pre-define a fixed number of line haul shipments (2, 3 or 4, respectively) as a long term exchange policy. The average performance of these policies are not insignificant (4.84%, 5.52% and 5.55%); however, when contrasted with the 7.06% average cost savings of the scenario (b), in which carriers make use of the proposed mechanism to dynamically define the allocation plan, we see a noteworthy difference. Two reasons can be identified to explain why a dynamic allocation plan outperformance the static exchange policies. The first one is that the variability of the LTL industry demands flexibility in the planning decisions in order define more effective operational plans. As a consequence, we see in the results that there is high variability on number of line haul shipments that the mechanism identified as the solution for every day (Figure 10). Therefore, a fixed number of line hauls policy represent a constraint for those days where the number of line haul shipments that are required to exploit collaboration benefits are either higher or lower than the pre-define number of line haul shipments(In the best case scenario, in which the fixed number of line haul trucks is 4, only 32% of the days this number corresponds to the actual efficient solution identified by the workload allocation mechanism as it is possible to see in the histogram labeled as figure 11).
Table 1 – Measurement of the quality of the solutions.
5.55% Porcentage of
Savings ( Quality of the Solutions
0 7.06% 4.84% 5.52% 1,232 16,109 616 13,687 14713 13983 Avareage total Kilometers 15,215 1284 15,220 (c) - 4 Avareage distribution routes total kilometers
Figure 10 – Representation of the daily optimal number of line hauls.
Figure 11 – Histogram of the optimal number of line hauls.
The second reason to understand the economic advantage of using the workload allocation mechanism proposed against a fixed policy lies on the content of the line haul trucks. Although in all the (c) scenarios the line haul trucks where filled using the same ranking list as the proposed mechanism, these shipments were loaded 100%. This practice is not necessarily efficient: a line haul shipment is an expense in which no order is delivered to the customers; hence, this operationalization of transshipments can only represent a source of reduction in operational costs if the distribution costs, which are the ones associated with the routes that deliver the orders to their final destination, are reduced enough to compensate the extra expenses of the line haul shipments. Accordingly, not measuring the impact of transshipping an order on the local distribution routes may lead to inefficient exchange of some customer orders, increasing the operational costs of the collaborative group. To illustrate this point, Figure 12 shows an example day in which the total kilometers travelled are depicted in function of the line haul shipments and the utilization of the last line haul truck. This figure shows how, for this specific day, the minimum costs (total kilometers) are achieved by utilizing the last line haul truck only at 80% of its capacity. This situation, as explained before, is caused by the fact that filling the additional 20% of the truck increase the total kilometers travelled by the distribution routes of the local carriers. When we replicate this analysis of the utilization of the last line haul truck for the complete simulation, we obtain the results that are shown in the histogram of figure 13. We can see in these results that in total, during the complete simulation, 66% of the times the last line haul truck was not filled completely.
0 2 4 6 8 10 1 9 17 25 33 41 49 57 65 73 81 89 97 10 5 11 3 12 1 12 9 13 7 14 5 15 3 16 1 16 9 17 7 18 5 19 3 20 1 Lh t ru cks Simulation Day 0% 5% 10% 15% 20% 25% 30% 0 1 2 3 4 5 6 7 8 Fr e q u e n cy
Furthermore, we can see that even filling up a line haul truck at a 20% of its capacity can be admissible under some conditions.
Figure 12 – Line haul shipments (and utilization of last truck) against Total kilometers.
Figure 13 – Histogram of the utilization of last line haul truck.
Finally, in order to conclude about the contrast between dynamic and fixed exchange policies, we identified that the correlation coefficient between the number of line haul shipments found by the mechanism and the total number of orders is approximately 0. This result lead us to conclude that there is no a systematic pattern (at least based on the number of orders) in which line haul decisions can be made, which highlights not only the importance of making workload allocation decisions dynamically, but also the need of using the proposed mechanism to maximize the benefit of collaborating.
7.3.2 Computational complexity performance:
The mechanism developed was proven to be very efficient in relation to computational efforts. None of the simulated days took longer than 120 seconds; even in the case where the number of orders was the highest in the data set (1100). As a result, we will not report further analysis on this aspect as it does not represent a drawback for this mechanism.
7.3.3 Robustness:
The solutions obtained through the complete simulation showed a variable pattern in relation to the cost savings achieved from day to day. Table 2 shows some descriptive statistics of the
12000 13000 14000 15000 16000 3 4 5 (60%) 5 (80%) 5 (100%) To taal km ilo m e te rs
Line haul shipments (with last truck utilization utilization)
Total kilometerstravelled by the line haul trucks
Total kilometers travelled in the distribution routes
Mean 7% Median 7% Standard Deviation 3% Range 15% Minimum 2% Maximum 17% Descriptive Statistics Cost savings of the
mechanism proposed
savings across the complete simulation. In these estimations we see that the potential benefits achieved through the mechanism range from 2% to 17%. In order gain insights on these differences from one day to another, we confronted some of the input parameters that changed throughout the simulation against the quality of the solution (total cost savings). Specifically, we graphically analyzed the quality performance of the mechanism against the total number of orders, the total number of loading meters and the average size per order (in loading meters). Although from day to day this factors change simultaneously, we see that the correlation coefficient between the loading meters and the number of orders is 0.79, which diminishes the negative effects of the simultaneous changes. Figure 14, 15, and 16 depict the performance of the quality in function of the considered inputs.
Table 2 – Descriptive statistics mechanism performance.
Figure 14 – Robustness of the quality of the solutions against the number of orders.
Figure 15 – Robustness of the quality of the solutions against the total loading meters.
0% 5% 10% 15% 20% 31 1 59 5 66 7 71 3 74 5 76 1 78 8 81 0 83 0 83 9 84 6 85 4 86 3 86 7 88 1 89 1 90 3 91 8 93 4 94 9 96 4 98 3 10 04 10 22 10 55 11 71 Por ce n tage o f c o st sav in gs Number of orders 0% 5% 10% 15% 20% 51 0 87 8 94 1 99 0 1,03 3 1,07 4 1,09 4 1,13 8 1,16 2 1,18 8 1,20 8 1,21 9 1,23 6 1,25 7 1,27 8 1,28 7 1,29 5 1,32 2 1,33 9 1,36 8 1,38 5 1,41 2 1,43 3 1,47 1 1,50 4 1,66 5 Por ce n tage o f c o st sav in gs
Figure 16 – Robustness of the quality of the solutions against the average size per order. The results show that there is no a clear pattern to which the quality of the solution is affected by any of these factors. This lack of relationship identified in the analysis is to some extent surprising. Considering the fact that throughout the simulation the market share of each of the companies was stable and balanced, it was expected that the greater the number of orders, the lower the cost reductions achieved since carriers would have been capable of making an efficient isolated plan due to economies of scale (as Cruijssen and Salomon (2004) suggest). Furthermore, it was also unanticipated that the average order size was not directly associated with the quality of the solutions, since we expected that the greater the average size per order, the smaller the number of orders that can be consolidated in the line haul shipment, which in many cases results in lower reductions in the costs of the distribution routes of the group. Therefore, we can argue that the geographical distribution of the final destinations of the deliveries is the critical factor that determines the total operation savings of the mechanism. We support this statement with the internal decision making process of the mechanism: the specialization of customer areas in order to avoid overlaps in the operation. Hence, not only its performance is highly dependent on the geographical overlap of the operations of both carriers, but also the quality of the isolated scenario is worse if more overlap in operations exists (which enhances the differences between the isolated and the collaborative scenarios). 7.4 Practical implications
The results obtained regarding the cost reductions that the carriers can obtain through collaboration should encourage the formation of collaborative carrier groups. We identified that collaboration among business divisions of a holding group is still economically beneficial when the costs of physical exchanges between carriers are considered. Although it was estimated that our approach can generate an average reduction in operational costs of 7.06% to the operation of the case study company, we cannot translate this number directly to the practical context since critical assumptions were made during the modeling phase of the research project. Nonetheless, analyzing the results from a qualitative perspective provides insights for workload allocation mechanism proposed and also for the problem that carriers are confronted with when performing collaborative planning. First of all, this research challenges a very common practice in LTL carriers’ collaborative groups, where carriers completely fill the line haul trailers in order to maximize their load factor. This research showed that a line haul truck not fully loaded is in some cases more beneficial than filling it
0% 5% 10% 15% 20% 1.3 55 1.4 16 1.6 39 1.3 69 1.1 96 1.4 43 1.4 34 1.6 48 1.3 70 1.3 72 1.2 62 1.4 98 1.4 78 1.1 76 1.3 71 1.2 58 1.5 03 1.5 51 1.5 26 1.4 01 1.2 10 1.4 09 1.4 14 1.3 77 1.6 17 1.3 67 Por ce n tage o f c o st sav in gs
complete. An explanation for this conclusion is the fact that the transshipment of orders has a direct repercussion on the local distribution routes and their costs. Therefore, including orders that have negative effect on the distribution costs is worse than leaving an empty space on the truck.
An additional insight that has been gained is the fact that collaborative planning decisions perform better if they are decided dynamically. The inherent variability and uncertainty of the LTL industry demands flexibility on the planning decisions from day to day in order to exploit collaboration. Accordingly, this research showed how the performance of a dynamic planning of the workload allocation mechanism outperforms fixed exchange policies that are common in practice. Nonetheless, we acknowledge that flexibility on the planning decisions also raises the complexity of the operational problems.
Finally, this research delivers an interesting mechanism through which carriers in collaborative groups can support their workload allocation decisions. The mechanism is convenient for such carriers since it provides, with little computational effort, a workload allocation plan on which carriers can base their decisions in order to reduce operational costs by specializing customer areas.
7.5 Theoretical implications
This research project addressed the collaborative planning problem for LTL intra-enterprise freight carriers. We considered the insights from previous allocation mechanisms in horizontal cooperation literature, and defined a gap in what has been proposed and the practical needs of LTL collaborative groups. This gap is grounded in the fact that none of the previous allocation mechanism for collaborative carrier groups considered the physical exchange of goods to balance their resources As a result, previous literature either disregarded the potential benefits of considering this physical exchanges (e.g., Krajewska and Kopfer, 2006; Berger and Bierwirth, 2009) or omit the inherent costs of transferring goods among partner carriers (e.g., Cruijssen and Salomon, 2004).
Considering the aforementioned gap, theoretical relevance of this research project is given by an estimation and evaluation of introducing transshipments in LTL intra-enterprise workload allocation mechanisms. Accordingly, the proposed mechanism obtained a reduction on the fulfillments costs of 7.03 % including transshipments and distribution costs (based on total kilometers travelled). This result not only establishes an interesting direction for the future development of workload allocation mechanism, but also shows that even when transshipment costs are considered, collaboration between LTL carriers can achieve transportation costs reductions of 5- 10%.
8. Conclusions and further research.
approaches, which are known as workload allocation mechanisms, attempt to minimize the total operational costs of a carrier group by allocating the transportation requests among the participating carriers. Although a number of mechanisms have been proposed in literature, there is still a gap between what has been proposed and the specific needs of collaborating carriers. This gap is explained by the two important considerations: the computational complexity of the models proposed in literature, and the lack of considering the physical exchange of goods between carriers in a collaborative group (transshipments). As a consequence of the latter remark, previous models either disregarded transshipments as a source of transportation cost, or omitted it as a form of collaboration. Consequently, this research project proposed a heuristic based workload allocation mechanism for an intra-enterprise context in which a holding company seeks to establish collaboration between two of its autonomous business divisions. The main rational of the approach designed is to exploit the opportunities that exist when there are overlaps in the transportation areas of the business divisions. Additionally, in this research project, transshipments are considered as a form of direct shipments between the facilities of the carriers (line haul shipments).
The proposed mechanism was tested by a simulation study considering real case study data. Results show that carriers, with overlapping customer areas, can achieve positive reductions in transportation costs by collaborating. Specifically, the performance of our approach in the case study generated on average 7.03% of savings in total kilometers travelled. However, since there is no benchmarking opportunity for the problem that is being dealt with, we addressed the theoretical relevance by showing the potential value of considering transshipments as a form of collaboration for future workload allocation mechanism in LTL carriers. On top of that, we showed that collaboration still generates economic benefits even when the transshipments costs are considered.
Furthermore, there are some considerations in relation to the workload allocation mechanism proposed that have implications on the numerical results of this thesis. The first one is that one of the assumptions of the model formulation is that pickups are not operationalized on the workload allocation mechanism. This assumption is grounded on the fact that line haul shipments of one operational day have no influence the line haul shipments of the same day. however, this implies that the numerical estimations in relation to the total costs are underestimated and the potential benefits of collaboration are not fully exploit since the pickups are not subject to be part of the collaboration. The second consideration is the fact the workload allocation mechanism was developed for LTL collaborative groups of carriers. Hence, the mechanism proposed cannot be implemented in contexts with more than two carriers are engaged in collaboration.
9. References:
Barr, R.S., Golden, B.L., Kelly, J.P., Resende, M.G C., Stewart, W.R. (1995). “Designing and reporting on computational experiments with heuristic methods”. Journal of Heuristics Vol.1, pp. 9-32.
Berbeglia, G., Cordeau, J-F., Gribkovskaia, I., Laporte, G. (2007), “Static pickup and delivery problems: a classification scheme and survey”. Top. Vol. 15 No. 1, pp. 1-31.
Berger, S. and Bierwirth, C. (2010). “Solutions to the request reassignment problem in collaborative carrier networks”. Transportation Research Part E. Vol.46, pp. 626-638. Bloos, M. and Kopfer, H. (2008). “Efficiency of Transport Collaboration Mechanisms”.
Communications of SIWN. Vol. 6, pp. 23-28.
Cruijssen, F. and Salomon, M. (2004). “Empirical study: Order sharing between
transportation companies may result in cost reductions between 5 to 15 percent”. CentER Discussion Paper 2004-80, Faculty of Economics and Business Administration, Tilburg University, The Netherlands.
Cruijssen, F. Cools, M., Dullaert, W. (2005). ‘’Horizontal cooperation in logistics: Opportunities and impediments’’. Transportation Research E: Logistics and Transportation Review. Vol.43, No.2, pp. 129-142.
Cruijssen, F., Cools, M., Fleuren, H.(2006). “Horizontal cooperation in transport and logistics: A literature review”. Transportation Journal. Vol.46, No.3, pp. 22-39.
Dondo, R. and Cerda, J. (2007). “A cluster based optimization approach for the multi depot heterogeneous fleet vehicle routing problem with time windows”. European Journal of Operation Research. Vol.179, No.3, pp.1478-1507.
Ergun, O., Kuyzu, C., Savelsbergh, M. (2007). “Shipper collaboration”. Computers and Operations Research. Vol. 34, pp. 1551-1560.
Ergun, O., Ozlem, O.O., Savelsbergh, M. (2011). “Lane exchange Mechanisms for truckload carrier collaboration”. Transportation Science. Vol.41, No.1, pp. 1-17.
Gomber, P., Schmidt, C., Weinhardt, C., (1999). “Auctions in electronic commerce: efficiency versus computational tractability”. In: Proceedings of the international conference on electronic commerce, 98, Seoul, pp. 43 – 48.
Hernández, S., Kalafatas, G., Peeta, S. (2008). “A less than a truck-load carrier collaboration planning problem under dynamic capacities”. Transportation Research. Vol.47, No.6, pp. 933-946.
Hernández, S. and Peeta, S. (2010). “Static single carrier collaboration problem for less-than-truckload carriers”. In Proceedings of the 89th Annual meeting of the Transportation Research Board, Washington, DC, pp. 44-75.
Krajewska, M.A., Kopfer, H., Laporte, G., Ropke, S., Zaccour, G. (2008). “Horizontal integration among freight carriers: Request allocation and profit sharing”. Journal of the Operational Research Society. Vol. 59, No. 11, pp. 1483-1491.
Krajewska, M.A. and Kopfter, H. (2006). “Transportation planning in freight forwarding enterprises: Request allocation and profit sharing”. OR Spectrum. Vol. 28, pp.301-317. Kopfer, H. and Wang, X.(2011). “Collaborative transportation of less than truckload
freight: request exchange through a route based combinatorial auction”. Chair of Logistics, University of Bremen, Bremen, September 2011.
Kopfer, H. and Wang X., (2011).”Increasing efficiency of freight carriers trough collaborative transport planning: Chance and challenges”. Available on:
http://www.sfb637.uni-bremen.de/pubdb/repository/SFB637-B9-11-005-IC.pdf (accessed12 of May, 2013).
Schwind, M., Gujo, O., Vykoukal, J. (2009). “A combinatorial intra-enterprise exchange for logistics services”. Information Systems and e-Business Management. Vol. 7. No.4, pp. 447-471.
Schonberger, J. (2005.). “Operational Freight Carrier Planning: Basic Concepts,
Optimization Models and Advanced Memetic Algorithms”. GOR Publications. Springer, Berlin.
