Article

 

Optimization in Supply Chain Management

by Dr. Heinrich Braun | SAPinsider

July 1, 2001

by Dr. Heinrich Braun, SAP AG SAPinsider - 2001 (Volume 2), July (Issue 3)
 

mySAP Supply Chain Management (mySAP SCM) comes with an integrated suite of optimizers - analytic functions based on state-of-the-art methods such as constraint programming, linear programming, genetic algorithms, and tabu search using best-in-class optimization libraries.

     With the mySAP SCM optimizers, you can see, on-screen and in graphical format, how to reduce production and transportation costs, increase through-put, deliver higher service levels, and ultimately improve product margins and your return on assets at all three planning levels (see Table 1). Optimizers are available at each level: strategic (long-term analysis and simulation), tactical (mid-term planning over the entire supply chain), and operational (short-term, day-to-day operations).

Level Optimizer Function
Strategic Supply Chain Design Optimizes the placement and capacities of plants, warehouses, and distribution centers in the supply chain using a suite of algorithms: metaheuristics based on Voronoi diagrams and mixed-integer linear programming. Equips companies to project ideal supply chain networks based on costs and profits, and to make outsourcing decisions.
Tactical Supply Planning Optimizes the global supply chain, from distribution centers to plants and suppliers. Automatically processes bills of materials while taking capacities into account, and optimizes transportation, production, and storage costs, along with revenues from demand. Uses linear or mixed-integer linear programming. Large-scale problems can be handled via several decomposition techniques (time, product, and priority decomposition).
Deployment Based on production quantities, this optimizer maximizes service levels while minimizing transportation costs. In a "fair-share" scenario (where there is less production than demand), high-priority demands are given preference. In a "push" scenario (where there is more production than demand), the range of coverage is balanced in the stocks of the warehouse in the supply chain.
Operational Transportation Planning and Vehicle Scheduling

Optimizes routes and schedules for pickup of orders at multiple depots, while considering multiple constraints: vehicle breaks, vehicle capacity constraints, and opening hours at depot and customer locations. It optimizes routes to minimize the total delivery cost, and can calculate costs using several dimensions (e.g., time, distance, stopoffs). By using appropriate cost dimensions, FTL/LTL costs can be modeled.

By setting penalties for early or late delivery, the optimizer takes into account requested arrival dates. Allows routing and dispatching of problems with several hundred orders. Includes a set of metaheuristics, such as tabu search, that guide local search methods and find optimized solutions for planning objectives. Heuristics use the runtime determined by the user and return the best solution found.

Production Planning and Detailed Scheduling A multi-objective optimizer that schedules orders according to manufacturing constraints. Handles complex environments with alternative routings and resources, secondary resources, and multi-stage production. Criteria such as service level (delay), setup time, setup costs, resource costs, and storage costs can be weighted in the objective function of the optimization. Users can customize the optimizer using several decomposition strategies based on one of two basic scheduling optimizers - the Constraint Programming (CP) or Genetic Algorithm (GA) optimizer - for flexible, fast, and efficient solutions, even in high-volume environments.
Campaign Optimization Optimizes the tradeoff between setup costs/times and inventory costs in manufacturing environments. Particularly suited to chemical, steel, and paper industries. Based on a two-phase approach: first, focusing on bottlenecked resources, the campaigns are optimized; second, optimizing the scheduling of these campaigns, taking into consideration all production levels.
Model Mix Optimization Allows production planners to determine the optimal order and scheduling sequence for manufacturing with a large number of variants. Various constraints are taken into account, including quantity and interval constraints, and assignment to a line or line segment. Especially suited to automotive and high-tech industries.

Table 1

mySAP SCM Optimizer for the Three Levels of Supply Chain Planning.

 

     Suppose your supply chain planner was looking at his planning table in SAP APO (Figure 1) and sees that the workload planned for June 26-30, indicated by the red box (all orders with a 30.06.2001 due date), does not actually fit into that time period. In fact, some orders start on the reactors the week before. He can improve the plan by:

  • Requiring that the earliest starting time of orders be 26.06.2001
  • Minimizing the due date delay (i.e., service level)
  • Minimizing the setup time and lead time (i.e., resource utilization)
Figure 1 Resource Planning in APO DS

     So he selects the "Optimizer" button in the toolbar to call the PP/DS optimizer. He then adjusts the parameter settings by selecting an optimizer engine ("Genetic Algorithm"),1 indicates the weighting factors for the objective function (e.g., factor "10" for setup and lead time, "20" for delay), specifies the al-lowed optimization runtime, and starts the optimization run (Figure 2).

     The "Status" information in Figure 2 reports on the progress of the optimization, indicating three steps:

  1. Reading the problem from liveCache.
  2. Searching (i.e., generating) a first solution.
  3. Improving the solution by continuing to optimize the solution until the runtime limit is reached.

Figure 2

Using the PP/DS Optimizer to Adjust the Supply Chain Model

     With the improved result shown in Figure 3, all orders fit now in the planned time period, even leaving some spare time on Friday afternoon (perhaps allowing some employees to leave a little early for the weekend!).

Figure 3

The New Planning Model, After Optimization

Supply Chain Support - No Matter How Complex the Planning Problem

Guided by a global objective function based on KPIs (for example, total setup costs, delay penalties, and transportation and production costs), optimizers generate and evaluate thousands of alternatives. If the optimization scenario is fairly straightforward - if, for instance, you request a linear optimiza-tion to be modeled in the Supply Planning component - it is solved optimally.

     In the case of highly complex planning scenarios, such as a production scheduling problem using alternative resources on several production stages, the optimizers are scalable to the runtime limit set by the supply chain planner. In other words, increasing the CPU time gradually improves the optimization result.

     For large, complex optimization problems, the runtime required to find the true optimum can be prohibitive. However, SAP's optimization suite gets around this by using several decomposition strategies. Decomposition "breaks down" the problem into smaller chunks, so the optimization suite can seamlessly bridge the gap between fast, rule-based, "greedy" heuristics and a truly optimized solution.

Limited Computing Resources Need Not Stand in the Way of Optimization

With all the benefits that mySAP SCM has to offer, supply chain optimization can still be limited by available computing power. mySAP SCM can compensate for less computing power by performing a more severe decomposition; however, this may decrease the quality of the generated solutions. To avoid this kind of degradation, mySAP SCM offers you the flexibility to address some of these issues at the source:

  • If limited computing resources are the result of multiple users working in parallel, this problem is alleviated within mySAP SCM itself, by a three-tier client/server architecture configurable for multiprocessor servers.
  • Separating the application servers can also help. We suggest running SAP's liveCache2 and the optimizer on separate servers. This architecture is especially recommended for modeling a large supply chain planning problem, since both liveCache and the optimizer require high-level computing power and high main-memory capacity.3
  • It is even possible to configure the system with several optimizer servers or with multi-processor architectures.

     Optimization and decomposition techniques offer great potential for parallelization, which optimizes separate parts of the supply chain in parallel sessions to reduce the load on the computing resources. With the multi-user capability of mySAP SCM, the production planner has the option to parallelize the optimization via a user-defined script.

     Moreover, with the most recent release of mySAP SCM, Release 3.0, the inherent parallelization of multiple agents may be used on a multi-processor server to solve large-scale optimization problems. This approach is already integrated into mySAP SCM decomposition techniques, as follows: For each decomposed planning problem, different agents may run in parallel. These agents may use different types of objective functions or basic optimizers. For a particular user-defined multi-criteria objective, we may define several agents by doubling, for each agent, the focus on a single criterion - one agent might double the setup costs focusing on resource utilization, while another doubles the delay costs focusing on service level. After completing such an optimization run using several agents, the user may select one solution from a set of solutions that have similar overall quality but that differ in a single criterion.

Customizing the Supply Chain Optimization Process

mySAP SCM's optimizer suite offers a generic solution for optimizing the supply chain that can be configured to master specific business scenarios. Because mySAP SCM optimizers have an open architecture, users can embed new basic optimizers and metaheuristics. Possible future enhancements include model functionality, improved optimization algorithms (libraries), and higher degrees of parallelization.

     The only limits of this algorithmic approach are computing power and the complexity of the supply chain itself. For large-scale supply chain problems, the user's challenge is to scale down planning complexity by taking advantage of the integrated solution for all three planning levels, as highlighted back in Table 1. The administrator's challenge is to maximize the enterprise's computing power to exploit the parallelism of a multiprocessor system. In the end, this dual approach will allow you to fully exploit the powerful capabilities of the mySAP SCM solution.


Dr. Heinrich Braun is the Development Manager for optimization algorithms in the mySAP Supply Chain Management solution. He is also a member of the computer science faculty at the University of Karlsruhe (independent lecturer), lecturing on combinatorial optimization problems. Prior to joining SAP, he led a research group on evolutionary algorithms and neural networks at the University of Karlsruhe from 1990 to 1996.


1 The basic scheduling optimizers are the CP (Constraint Programming) or GA (Genetic Algorithm) approaches. The CP approach is a more generic scheduling optimizer, while the GA scheduler is based on an evolutionary approach for generating a new supply chain solution.

2 liveCache is an SAP object database in main memory, designed especially for planning problems with vast amounts of data. It is included in SNP, PP/DS, ATP, and TP/VS modules as of mySAP SCM, Release 3.0.

3 However, if absolutely necessary, smaller companies can use a single application server for both liveCache and the optimizer.

An email has been sent to:






More from SAPinsider



COMMENTS

Please log in to post a comment.

No comments have been submitted on this article. Be the first to comment!