OPTIMAL ACQUISITION POLICY IN REMANUFACTURING SYSTEMS WITH QUANTITY DISCOUNT AND CARBON TAX SCHEME

Original scientific paper High uncertainty and variability of acquirable cores can lead to some difficulties of operations management in the remanufacturing process. Two acquisition models with quantity discount and carbon tax scheme were established to analyze the effect of quality variability and condition uncertainty and to reduce these difficulties. In the first model, quality variability was considered. The optimal acquisition policy was determined by a single closed-form expression because the average cost of remanufactured products is independent of the acquisition quantity for a given discount segment. In the second model, both quality variability and condition uncertainty were considered. The properties of the model were analyzed, and a single algorithm was developed to obtain the optimal acquisition policy based on the bisection method. Finally, numerical simulation examples were provided to explore the effect of quantity discount on the optimal acquisition policy. Results show that the quantity discount schedule impels remanufacturers to acquire more cores and enhances the chances of remanufacturers to select the best cores. Moreover, the total expected cost decreases, especially in the case with both quality variability and condition uncertainty. The proposed model can effectively solve the acquisition problem in remanufacturing systems.


Introduction
With respect to the current competitive business situation and the environmental protection concern, remanufacturing is an effective way to maximize value creation in the entire product life cycle and reduce the emissions of carbon dioxide (CO 2 ) and other heattrapping gases [1,2].In practice, firms develop remanufacturing businesses to obtain social, economic, and environmental benefits [3,4].
Core acquisition has been considered as the key input to assess the potential economic attractiveness of remanufacturing activities.During acquisition, high uncertainty and variability of acquirable cores (i.e.quantity, quality and time) are crucial features that pose great challenges to remanufacturers [6,7].Specifically, uncontrollable return makes the production planning and control of remanufacturing systems difficult [8÷11].Moreover, the remanufacturer can increase its selectivity as more cores are acquired and select the best cores to reduce remanufacturing costs [9].The remanufacturer is motivated to take measures to collect more cores.Quantity discount, as a frequently used promotion policy, encourages brokers or collectors to provide more cores.
In addition, environmental burdens (i.e.energy consumption, material consumption, and carbon emission) have significant effects on remanufacturing systems.The remanufacturer should incorporate environmental factors into remanufacturing decisions [12].Thus, the remanufacturer should consider the tradeoff of between environmental and economic benefits when acquiring cores.According to the above analysis, this paper examines a core acquisition problem on the highly variable quality of acquirable cores.Subsequently, the effects of quantity discount and environmental burdens are considered.
The rest of this paper is organized as follows.In Section 2, the related literature review is discussed, and the differences between our paper and the literature are analyzed.In Section 3, the model with quality variability is developed using a nonlinear mixed integer programming model.The optimal solutions are analyzed and derived.The extension model with both condition uncertainty and the quality variability is also developed in this section.In Section 4, numerical simulation examples are provided to illustrate the model and the approach with managerial insights.Section 5 concludes the paper with future research directions.

State of the art
Core acquisition, as one of the key steps in remanufacturing, has attracted considerable attention from researchers.A number of researchers have discussed various optimal operational policies in core acquisition, and proposed that positive core acquisition could influence or control the quality and quantity of acquirable cores [13÷19].
Scholars have focused on the effect of core quality on the optimal policies of core acquisition owing to the inherent variability of acquirable cores.The effect is mainly discussed from the viewpoint of both condition uncertainty and quality variability.
The number of cores to be remanufactured in the uncertainty condition is stochastic for a certain acquisition quantity in core acquisition (i.e., the proportional yield is stochastic).Some researchers have discussed how core acquisition is affected by condition uncertainty.For example, Zhou and Yu studied the optimal acquisition policy in a reverse supply chain, in which the proportional yields are uncertain at two possible correlated collection sites [20].Zikopoulos and Tagaras analyzed the optimal quantity of core procurement and remanufacturing in a two-stage reverse supply chain, and discussed the optimal application condition of quick sorting [21].Van Wassenhove and Zikopoulos highlighted the effect of quality overestimation on the optimal acquisition decisions under the deterministic and stochastic demand [2].Li et al. examined a coordination issue between acquisition and production planning, in which the proportional yield is stochastic and the core return is sensitive to the acquisition price in a hybrid manufacturing and remanufacturing system [22].These studies assumed that the acquirable cores had been sorted, and each grade of cores was homogenous, that is, quality variability was not considered.
Quality variability implies that the quality of each acquirable core is assumed heterogeneous in a core acquisition.Some researchers have analyzed how core acquisition was influenced by quality variability.Galbreth and Blackburn developed an acquisition and sorting model with quality variability under both deterministic and uncertain demands [23].Lu extended this model to two cases: the acquisition cost with a piecewise linear convex function and with a multi-period decision issue [24].Ferguson et al. examined a multi-period production planning problem facing different quality levels of cores [25].These studies investigated the influence of quality variability.However, core condition is deterministic, that is, the proportional yield is deterministic.Some researchers have also discussed the effect of both condition uncertainty and quality variability on the decision process of core acquisition and remanufacturing [26÷28].Detailed discussions and reviews can be obtained from [28,29].However, the environmental burdens in the remanufacturing process were not considered in the aforementioned papers.
Krikke studied the effect of carbon footprints on the design of a closed-loop network [30].Kannan et al. developed a similar model [31].Liu et al. developed distribution-free newsvendor models to determine the remanufacturing policies under three carbon emission regulations [3].Yang et al. examined a multi-product acquisition and remanufacturing problem under a variable core quality and carbon tax scheme [12].However, quantity discount was ignored in the aforementioned papers.
In the current paper, quantity discount is incorporated into the core acquisition problem.Moreover, carbon emission is a classic environmental burden.The effect of carbon emission is incorporated into the problem based on the carbon tax scheme.First, the problem, which considers quality variability, is studied and then extended under both condition uncertainty and quality variability.In the two cases, the problem is formulated as nonlinear mixed integer programming models.The properties of the models are analyzed, and the optimal solutions are derived by the effective approaches.

Methodology 3.1 Problem descriptions and assumptions
In this paper, the remanufacturer is assumed to acquire cores from third-party brokers or collectors in a market-driven remanufacturing system.Facing a deterministic demand m, the remanufacturer determines the acquisition quantity n (n ≥ m) according to the quality of acquirable cores (shown in Fig. 2).
The remanufacturer should select the suitable indices and estimate the quality using historical data because the quality of each acquirable core (t) is different and unknown before acquisition (i.e., quality variability).t is usually measured by a continuous indicator, such as total running time in the entire machine lifetime [32], remanufacturing cost [23,24], and remanufacturing process time (i.e., time required for remanufacturing) [33].The higher the value of t is, the worse the quality.Here, t is a random variable with the cumulative distribution function of ( ) G ⋅ and the probability density function of ( ) g ⋅ .According to Galbreth and Blackburn, for a given t, the unit remanufacturing cost is a ct + , where a is the unit fixed cost, and c is a variable component that depends on core quality [26].
The remanufacturer provides a price schedule with price breaks at quantities (q i ) to encourage brokers or collectors to offer adequate cores, where In addition, many counties and economic communities have implemented a series of regulatory mechanisms to reduce carbon emission [34,35].The carbon tax scheme is a frequently used cost-effective measure.This paper incorporates carbon emission into the model under the carbon tax scheme.
After the cores are purchased, the remanufacturer selects the best m cores to be remanufactured.The remaining n m − cores are scrapped because of poor quality with a unit scrapping costs.The notation used in this paper is as follows:

Model with quality variability 3.2.1 Model formulation
When the quality of the acquirable cores is estimated, the remanufacturer provides a threshold value (t 0 ) and analyzes which acquirable core should be remanufactured or scrapped [23].That is, the cores of high quality (i.e., t≤t 0 ) are remanufactured and the cores in poor quality (i.e., t>t 0 ) are scrapped (shown in Fig. 3).
Accordingly, the expected remanufacturing rate is ).For the given . Here, the average remanufacturing cost is: .)d ( The remanufacturer minimizes the total cost, which includes the acquisition, remanufacturing, scrapped, and carbon emission costs, to derive the optimal acquisition quantity.When the cores are acquired at the discount segment i ( 1, 2, , i l =  ), the total cost is given as In Eq. ( 1), the first term is the acquisition cost, the second term is the scrapping cost, and the last term is the remanufacturing cost.Based on the above analysis, the model with quantity discount is given as: , , In P1, Eq. ( 3) limits the upper bound of the acquisition quantity at the discount segment i, while Eq. ( 4) restricts the lower bound of the acquisition quantity at the discount segment i.Both ( 3) and ( 4) can be used to ensure the availability of discount segment i because y i is a binary variable.Eq. ( 5) ensures the uniqueness of the discount segment.Eq. ( 6) means that the deterministic demand should be satisfied.Eq. ( 7) represents the integer constraints.The last constraints in Eq. ( 8) are binary constraints.

Solution
P1 is a mixed integer nonlinear programming problem.The properties of the model are further analyzed to solve the model effectively. As Tc n can be transformed as 0, 0, where, From Eq. ( 9), Ac(u i , t 0,i ) is easily identified as independent of m [12,23].Ac(u i , t 0,i ) can be viewed as the average cost of remanufactured products in discount segment i. Here, the decision variable of Tc i (n i ) is changed from the acquisition quantity i n to the threshold value t 0,i .
As shown above, the optimal solution of P1 can be derived in two steps.

Let
As the left-hand side of Eq. (10) Here, the total cost is calculated as follows: * 0, ( ) ( ) Tc n m a c e ct c e s P1 can be viewed as the special case of [24].The relationship between * i , t 0 and * t 0 is expressed as follows: Eq. ( 12) can be simplified because of the structure of the discount schedule in P1.
( ) G ⋅ is the cumulative distribution function.From Eq. ( 10), we have: Tc n are increasing functions of i u because of the monotonicity of ( ) i G ⋅ and Eq. ( 11).
Thus, when does not exist.Thus, Eq. ( 12) can be simplified, and the optimal threshold value * 0 t is derived as follows: Step 2: Determining the optimal acquisition quantity n ** .After * 0 t is derived, n ** can be determined through the expression ** * 0 / ( ) n m G t = .

Model with both quality variability and condition uncertainty
Section 3.2 only considers quality variability.That is, the remanufacturing rate G(t 0 ) is deterministic, and the average cost of remanufactured products 0, ( , ) i i Ac u t is independent of the acquisition quantity.
In this section, the model is extended to the case with both quality variability and condition uncertainty.Considering the condition uncertainty of acquirable cores, the remanufacturing rate G(t 0 ) is stochastic.Furthermore, the higher the acquisition quantity is, the greater the choice of cores with higher quality.Therefore, the remanufacturing cost of the best m cores is lower, that is, the average remanufacturing cost is related to the acquisition quantity.Here, let X (1) , X (2) , …, X (n) denote the ascending order statistics of the n cores, ordered by quality, and the corresponding quality are t (1) , t (2) , …, t (n) , and ( 2) ( ) (1) Thus, the remanufacturing cost of the best m cores is expressed by In addition, the other problem descriptions and assumptions of this model are the same as those in Section 3.2.
When the cores are acquired at the discount segment i In Eq. ( 15), the first term is the acquisition cost, the second term is the scrapping cost, and the last term is the remanufacturing cost.Accordingly, the complete model with both condition uncertainty and quality variability is given as: P2 is also a mixed integer nonlinear programming problem.The properties of the model are analyzed to solve the model effectively.
First, the following model can be derived for discount n ** is the optimal solution of P2, and * i n is the optimal solution of P3.Then, the minimal objective function value of P2 is ** * arg min{ ( ), 1, 2, , } Let i n + be the optimal solution of P3 without constraints (18)(19).
Carbon tax changes the remanufacturer's costs, but the properties of the problem are still similar to [28].
Thus, i n + can be derived on the basis of [28]. ( Furthermore, Yang et al. proved that Tc i (n i ) is a strictly discrete convex function of n i (n i ≥ m) [28].Based on the above research results, the relationship between i n + and * i n is analyzed as follows: Proposition 1 Proof.
(a) The statement holds straightforwardly.(b) ( ) Tc n is a strictly discrete convex function of i n Tc n monotonically decreases for Tc n monotonically increases for i n According to Proposition 1, the optimal solution of P2 (i.e., ** n ) can be derived by the following algorithm: Algorithm 1: Step 1. Initialization: starting from the lowest unit acquisition cost, i.e., i l = .
Step 2. If 0 i = , let 1 i = , and go to Step 5; else, calculate i n + by a single bisection method.
Step 3. Check the discount bound constraints: If , and go to Step 5.
Let 1 i i = − , and go to Step 2.
Step 5. Compare the objective values of In Step 2, i n + is calculated by a single bisection method from [28].Solving P3 is unnecessary for all discount segments in the above algorithm.Without loss of generality, Algorithm 1 will stop in discount segment j once 1 j j j q n q + − ≤ < . As Tc n Tc n Tc n Tc n Tc n Tc n Tc n Therefore, we only calculate the objective values of * i n with i j ≥ in Step 3 or 4. When the minimal value is selected in Step 5, the corresponding solution is optimal to P2.In addition, Algorithm 1 can be applied to any continuous quality distribution because the quality distribution is not hypothesized.[28].A third-party broker or collector is assumed to offer three discount segments: (1) smaller than 200, (2) between 200 and 299, and (3) larger than 300 with unit acquisition costs of 2,80, 2,65 and 2,50, respectively.Other parameters are set to the following: s = 0, a = 0, c = 8, e = 0,1, e s = 0,2, c e = 1.
The case without quantity discount is considered to analyze the effects of quantity discount.In non-discount case, u = 2,8.The results in the numerical examples are analyzed under frequently used quality distributions (i.e.uniform, exponential, and Weibull distributions).The means of the different distributions are assumed to be 2.

Effect of quantity discount
For the model with quality variability, the optimal results are shown in Tabs. 1 and 2.
In Tab. 1, the optimal acquisition threshold value of the given discount segment ( * 0,i t ) is derived by Eq. ( 10) under three distributions.Accordingly, the average costs of the remanufactured products (Ac(u i , t 0,i )), and the remanufacturing rate ( * 0, ( ) i G t ) are calculated.Based on the above results, the optimal threshold value ( * 0 t ) is derived by Eq. ( 14).The optimal acquisition quantities (n ** ) and the total costs are calculated under different demands (shown in Tab. 2).
Tab. 2 compares the optimal acquisition policy under the discount and non-discount cases.In the discount case, the remanufacturer can acquire more cores and select more high-quality cores because unit acquisition cost decreases, thus resulting in the decrease in the threshold value and the remanufacturing ratio.Accordingly, the average cost of remanufactured products and the total cost decrease (e.g. the optimal results in the Weibull distribution).Tab. 2 illustrates that when only quality variability is considered, the threshold value may decrease by up to 7,49 % ((0,8436-0,7804)/0,8436), the acquisition quantity may increase by up to 2,35 % ((341−333)/341), and the total cost may decrease by up to 7,59.% ((1330−1229)/1330).The total cost decreases by 2,51.% on average.
For the model with both the quality variability and the condition uncertainty, the optimal results under different distributions are shown in Figs.4÷6.
According to Tab. 2 and Figs 4-6, the remanufacturer can similarly acquire more cores in the case with both the quality variability and the condition uncertainty.Unlike in the case with the quality variability, the remanufacturing ratio is no longer decreasing monotonously.Given the demand (m), the optimal acquisition quantity may be more than that in the case with quality variability, and the total cost declines (e.g., the optimal results in the Weibull distribution).Fig. 6 illustrates that the acquisition quantity may increase by up to 2,34 % ((342−334)/342), and that the total cost will decrease by up to 7,57.% ((1334−1233)/1334).The total cost decreases by 3,63.% on average.The nonlinear mixed integer programming models were established under the quantity discount and carbon tax scheme to reduce the operation and management difficulties caused by high uncertainty and variability of acquirable cores.The following conclusions were obtained: (1) When quality variability is considered, the average cost of the remanufactured products is independent of the acquisition quantity for the given discount segment.The optimal acquisition solution is derived in the closed-form expression.
(2) When both the quality variability and the condition uncertainty are considered, the total expected cost is strictly a discrete convex function of acquisition quantity for the given discount segment.The optimal acquisition solution is derived on the basis of the bisection method.
(3) The case studies considered in current paper verify that the quantity discount schedule can encourage remanufacturers to acquire more cores and effectively reduce the total expected cost, especially in the case with both the quality variability and the condition uncertainty.
This study can be extended to the case with stochastic demand.Moreover, the multi-product acquisition problem with quantity discount is worth studying.

Figure 1
Figure 1 Typical remanufacturing process

Figure 2
Figure 2 Process of acquiring and sorting cores choose the best as the solution of P2.That is,

Figure 4
Figure 4 Optimal results under the uniform distribution

Figure 5 Figure 6
Figure 5 Optimal results under the exponential distribution be the minimal solution of Tc i (n i ).On the basis of the first order condition of Tc i (n i ), we can derive * 0,i t , ∫

Table 1
Optimal results of the model with quality variability (I)

Table 2
Optimal results of the model with quality variability (II)