Other techniques such as frequency response plot, white noise amplification, noise bandwidth [58] and worst-case amplification were used to quantify the variability in order and inventory. In recycle supply chains, the product recovery for re-manufacturing has both environmental and commercial benefits. The impact of remanufacturing rate and their lead times on inventory variance and the bullwhip of the manufacturing system were investigated using a derived hybrid dynamic model of a manufacturing-remanufacturing system [59]. Larger return rate of used stock for remanufacturing reduces the inventory variances and bullwhip in the manufacturing plants. The longer in-use time and large remanufacturing lead time have less impact on inventory variance and bullwhip reduction in new prod-uct prodprod-uction system. Prior researches conclude that bullwhip effect is inevitable in the distribution system practicing order-upto-level policy. Other replenishment policies are prone to bullwhip effect because of inaccurate demand forecasting, lead time and inappropriate replenishment rule parameters.
2.3 Distribution System
Mathematical models of distribution system has helped various researchers to study the dynamics of distribution logistics with a view to improve their per-formance [60, 61]. The deviation in performance of production-distribution sys-tem from optimal behavior is identified in the three-echelon Forrester production-distribution system [47]. Five different approaches were proposed to improve the behavior of the system. Decision policies related to replenishing product for in-ventory allocation play a crucial role in the nature (aggressive or conservative) of
2.3 Distribution System
distribution nodes. Hybrid dynamic simulation tools were developed to analyze the impact of several heuristic decision-making policies (Make-to-stock, Make-to-order and true-order policy) on the dynamic behavior of a supply chain system [62]. This approach was extended to a single manufacturing site, multi-product processing unit with the distribution network consisting of a warehouse, a distribution center and a retailer with three different customers [63]. Several ordering policies were analyzed to optimize the decision-making rules with respect to performance indices like supply chain cost, customer satisfaction and demand amplification factor. A proper replenishment rule that results in minimum bullwhip and inventory vari-ance is identified by utilizing the analytical expression derived to quantify the bull-whip and variance in inventory position [56]. A new replenishment rule called the smoothing order policy was proposed to overcome the bullwhip effect by generat-ing smooth order patterns. This was achieved by considergenerat-ing the error of inventory at hand and on road separately rather than together. By applying linear control theory, the stability and performance characteristics of inventory on-hand policy and base-stock policy were studied [57]. Recycling the used product has both en-vironmental and commercial benefits. Closed loop supply chain recycles the used product to reuse purposes to reduce waste disposal, refabrication cost and time.
A hybrid dynamic model was developed for a manufacturing-remanufacturing sys-tem to understand the benefits of the production rate of remanufacturing firm and their positive influence on the manufacturing systems [59].
Another aspect of supply chain representation is surrogate models [64]. The sur-rogate models are constructed from the real data (input and output) using least
2.3 Distribution System
square support vector machine (LSSVM). Capturing even very complex input and output relations by discriminating the true signal from the noisy data is a remark-able property of this approach. The essential idea is to fit a single surface for the whole solution space, and use this surface to perform optimization instead of the simulation model. The complexity of the model and the convergence of the process to a local or global optimum depend on the type of surrogate used. The main challenge is in developing an accurate surrogate model that is adequate over the complete design space. The useful insights about the global behavior of the system and less computation burden are the highlight of using surrogate model.
2.3.1 Modeling & Control of the Distribution Node
The decentralized distribution node is represented using a discrete system based on material and information balances. Chapter 3 contains the detailed repre-sentation of distribution node model, which helps to analyze the stability and performance of heuristic replenishment rules under batch replenishment [65] and continuous replenishment strategy [18]. The derived model was utilized to ana-lyze the performance of the distribution node under various conditions such as:
(a) infinite supply and high stock in the upstream and downstream nodes (b) infinite supply and low stock in downstream node and (c) limited supply from upstream nodes. The heuristic replenishment rules namely the order-upto-policy, proportional-integral (PI) and cascade control were analyzed with respect to the performance metrics of excess inventory, backorder, inventory discrepancy and de-mand distortion. These studies revealed that the PI and cascade policies provides
2.3 Distribution System
less backorder at the expense of holding higher excess inventory, whereas order-upto-policy provides less excess inventory at the cost of high backorder. In batch ordering strategy, bullwhip effect is inevitable for the standard heuristic order-ing policies with demand forecaster. Irrespective of the demand forecaster, the order-upto-policy produces stronger bullwhip effect due to large lead time under continuous ordering conditions. For the distribution system facing a stochastic de-mand pattern, the minimum variance control (MVC) strategy was derived for the distribution system by characterizing the demand pattern [19]. The MVC strategy was obtained with the objective of minimizing the errors between the predicted inventory level in-hand and on-road and the desired targets. By characterizing the downstream order as an ARIMA model, the MVC strategy uses the ARIMA model information to predict the future inventory level at-hand and on-road and act accordingly to reduce the errors in the inventory-at hand and on-road. The performance of the distribution node was described as the function of backorder and excess inventory with the bullwhip constraint. For both stationary and non-stationary demand trends, the optimized performance of the MVC was found to be superior to the optimized performance of the other existing schemes such as order-upto policy, proportional-integral control and smoothing order policy with bullwhip constraint.
2.3.2 Distribution Network
Recent research activities have focused on developing supply chain network mod-els to simulate various strategies to attain better control of the network behavior
2.3 Distribution System
through simulation based optimization techniques. Ko et al [66] developed a hy-brid optimization based simulation model through genetic algorithm. This study aims to achieve an efficient (dynamic) distribution network facing uncertainties in demand, transportation and processing time. As a result of the dynamic environ-ment, supply chains are operated using third-party logistics providers (3PLs) to make a sequence of inter-related decisions over time. Decisions such as opening or closing of facilities at various warehouses in the network, amount of material flow from a client to a warehouse and to the market and capabilities of ware-houses are made by the 3PLs. Braun et al [67] adopted the model predictive control (MPC) technology to supply chain management. Their proposed strategy performed reasonably well in tracking realistic demand despite of uncertainties in production lead time and demand forecaster. The application of this strategy to a three-echelon, six node system confirms low safety stock requirement compared to other heuristics. This strategy handles the uncertainty in plant-model mismatch, information sharing and constraints in an effective manner. The supply chain system with large lead time requires both feedforward and feedback action for better management. The combination of information sharing to provide feedfor-ward action and move suppression to temper feedback action offers huge benefits in reducing inventory levels, smoothing the order patterns and inventory fluctua-tions to attain better customer satisfaction and enhanced performance. Rasku et al [35] compared model reference control with model predictive control to investi-gate their capability in minimizing variations in inventory level of the supply chain.
The penalization function defined in the MPC strategy causes difficulty in tuning the penalization parameter to prevent excess orders. The tunable parameters were