Step 1: Identify the System Constraint

In this case, we are dealing with a situation in which the constraint is the available capacity at a resource. The simplest way to identify such a constraint is to compare the load placed on each resource with the total amount of production and setup required at that resource to satisfy the market demand. However, this does not always produce meaningful results due to inaccuracies in data. In several hundred factories with which the author has consulted, this method fails to identify the real bottleneck in an overwhelming number of cases. Detailed procedures for identifying the bottleneck have been developed for each of the different production flows-V, A, T, and I-and are briefly discussed at the end of this chapter. (See Srikanth and Umble, 1997, Vol. 2, Chapter 4 for V-plants, Chapter 5 for A-plants, and Chapter 6 for T-plants.) The choice of the bottleneck is the pivotal point in the development of the strategy for the entire business and hence this is a decision that must be made by the business as a whole and is not just a production/manufacturing decision.

Step 2: Decide How to Exploit the System Constraint

The constraint in the environment we are discussing is the available capacity at a specific resource. Exploitation of that resource means that we should maximize performance with respect to the global operational metrics of Throughput, Inventory, and Operating Expense. More specifically, the goal is to maximize Throughput, while efficiently managing Inventory and Operating Expense. How can we maximize the Throughput of the production operation with a specific capacity constraint or bottleneck? To answer this question, we can look at ways in which capacity is currently wasted.

By definition, the load placed by current market demand on a bottleneck is greater than or equal to the available capacity at this resource. If the resource spends any time doing something other than what is required for current market demand, then Throughput will be negatively impacted and we will not have properly exploited the available capacity. It is, therefore, critical that every item produced at the constraint be a product that is required to fulfill short-term market demand. Another way in which capacity at the constraint can be wasted is for the resource to suffer a breakdown and then a significant time to elapse between the resource breakdown and its being fully operational. Excessive setup times, time lost during shift changes or lunch breaks, etc., are all ways in which capacity at the bottleneck is wasted and represents the opposite of exploitation. Policies, such as overlapping shifts and staggering breaks should be put in place to eliminate these forms of wasted capacity. Capacity is also wasted when the bottleneck works on products that are not needed to satisfy current market demand. While this may appear to be so obvious as to constitute a triviality, the reality in most operations is very different. Motivated by local optima considerations, often bottleneck resources end up working precisely in this wasteful way-because either no other work is available or batch sizes in use are excessive. One of the prime considerations in designing the rules that will allow proper exploitation of the bottleneck is to make sure that the bottleneck does not run out of work and that the planned work (as well as actual material available on the factory floor) consists only of the product required to meet very near term demand. The procedures for doing this are discussed in the section on the drum.

Due to the existence of dependencies in manufacturing operations, the performance of any resource is influenced by the performance of other resources. In the simple flow shown in Fig. 8-2, resource R4 cannot continue to work if resource R2 is down for an extended period of time. If resource R4 is a non-bottleneck, then the forced downtime at R4 is not a serious issue. If, however, R4 is the bottleneck in this production flow, then the downtime at R4 is unacceptable as system Throughput is reduced. To ensure that resource R4 can continue to work even when upstream flows experience disruptions, we must maintain a buffer at R4 (enough work to cover the time the resource is down). Since the objective of this buffer is to protect R4 from upstream disruptions, the size of this buffer is a function of the magnitude and frequency of these disruptions. While determining the "optimum" buffer size is quite complex, the two limits are obvious-the buffer should not be so small that the bottleneck is frequently at risk of running out of work and the buffer should not be so big that the total lead time for the flow is excessive. In the section on buffers, we discuss the procedure for how to set the size of these buffers.

Step 3: Subordinate Everything Else to the above Decision

Once the bottleneck or capacity constraint has been identified, policies to ensure full productive utilization is put in place, the resource properly buffered, and the planned flow through this resource is identified, then Step 2 is complete. The next step, Subordinate, is to make sure that all other resources focus on performing tasks in such a way that the planned flow through the constraint is supported.

All activities from the release of material to how they are processed upstream and downstream of the bottleneck should be done in a manner that best supports the decisions made in Step 2. It is important to realize that while the discussion on the constraint is powerful and interesting, the task of execution falls mostly on Step 3 and the management of non-constraints. This is a simple consequence of the fact that most resources (95 to 100 percent) are non-bottleneck resources and to control execution means controlling what is happening at these resources.

The subordination required by Step 3 is made challenging because the mentality fostered by traditional cost world management is not consistent with subordination. This is the point at which the third principle of flow management (efficiencies must be abolished) needs to be implemented. The case in Fig. 8-4 illustrates this point. Resource R2 is the constraint and can process 100 units per day. Resource R1 is a non-constraint and can process 120 units per day. Subordination requires that R1 only process 100 units per day, but traditional mindset would encourage R1 to work to its full potential and thus produce in excess of 100 units per day. In almost every implementation of DBR that the author has done, the task of subordinating (or holding back production) non-constrained resources has been the most challenging and difficult task.

FIGURE 8-4 PFD for a one-product flow line indicating the production capacity of the different resources.

An alternate way to look at Step 3 is as follows. Steps 1 and 2 have established the total flow that must be achieved-product mix, volumes, etc. In accordance with the section on managing flow, we must now implement the four principles of flow. In particular, recognizing that improving flow is the primary objective, we must establish how to implement Principle 2-a mechanism to prevent overproduction. This overproduction (Sugimori et al., 1977) is the first and most important waste that is explicitly identified in TPS as well as in JIT, Lean, and other offshoots of TPS.

The process by which subordination is enforced in the DBR system is the rope and is discussed in a later section.

Steps 4 and 5

At the completion of Step 3 (subordination) we have a system that is operating at full potential-we are getting the maximum Throughput by what has been done at the constraint and waste is minimized by subordination at all other resources. In order to improve the performance of the system further yet, we must raise the performance of the constraint itself. However, when the performance capability of the current constraint is elevated, it may no longer remain the constraint. Its new potential may be larger than the capability of another resource in the system. Steps 4 and 5 are designed to deal with this possibility. Since our focus is managing a plant that has a current constraint, we will not discuss ramifications of Steps 4 and 5. Rather, we proceed to the implementation of Steps 2 and 3 through the DBR mechanisms.

The DBR System

We now discuss the specific procedures and methods that make up the DBR system for planning the flow of product through manufacturing operations. BM is the execution control portion of the DBR system. The objective of the DBR system, as for any planning and control system, is to meet Throughput expectations while efficiently managing Inventory and Operating Expense.5 The essence of the DBR approach is captured in Fig. 8-5.

The Drum

The drum considers the constraints in the system and firm customer commitments, in setting the pace for the entire system. The process of setting the drum begins by identifying the work that needs to be done at the constraint by the total output required. In the case of companies that are make-to-order (MTO), this is the work required to be performed by the CCR to meet all customer requirements that fall in a given time period (for example, all orders with customer due dates in the next 30 days). In the case of make-to-availability (MTA)6 companies, the output requirement is the total finished products required to fill the stock buffers. Once we have a list of what must be produced at the constraint, it is then simply a matter of determining the sequence of production (which product first, which product next, and so on) and the production batch size (how much will be produced once we start a specific product). Factors that should be considered in deciding the production sequence and the size of the process batch as well as detailed examples are found in Srikanth and Umble (1997, Vol. 1, Chapters 7 and 8) and Schragenheim and Dettmer (2001).

FIGURE 8-5 Illustration of the basic DBR system.

The Buffer

In a world free from disruptions, such as resource breakdowns, process yields, etc., the production lead time-the time that we allow for the raw material to be transformed to a finished part or product-can be simply equal to the sum of the process times and setup times at each step of the routing for that product. In the real world where there are many forms of disruption, the use of a planned production lead time equal to the sum of processing and setup times would be considered foolish and rightfully so. Any disruption such as a resource breakdown would make it impossible to produce the product on time. The actual production lead time will always be larger than the sum of process times and setup times. Since disruptions are unavoidable, planned lead times will have to be larger than the sum of process and setup times. This is true if we are to have any chance of making the actual production lead time equal the planned production lead time.

Whenever there is a task that is subject to variability, it is clear that the actual time the task is executed-started or finished-is going to be different from any plan that does not allow some degree of padding in the form of safety time. This is essentially the concept of the time buffer.7 What makes the application of the time buffer concept unique and powerful is the explicit recognition that the goal of a DBR planning system is not to make each task to be on time to a planned schedule, but to make the actual flow through the system sufficiently reliable to satisfy market demand. In other words, the objective is not to protect the ability of each task to be on time (to a plan) but only to make sure that the entire system is on time. This recognition allows us to provide a significantly higher degree of reliability in a DBR plan than one that tries to ensure protection for each step in the process (as in a push system or Kanban pull system). In addition, this higher degree of reliability can be accomplished at a significantly lower production lead time.

Specifically, a time buffer is defined as follows. A time buffer represents the additional lead time allowed, beyond the required setup and process times, for materials to flow between two specified points in the product flow. Two points8 commonly used in this context are material release (gating operations) and receipt of a finished product at a warehouse (MTA) or at shipping (MTO). The objective of these time buffers is to protect the system Throughput from the internal disruptions that are inherent in any process. The relationship between production lead time and process times can be expressed by the following relationship.

Production lead time = Sum of process times and setup times + Time buffers The concept of time buffers is almost self-evident. Determining the proper size of a time buffer on the other hand appears to be a complex task. Since the objective of the time buffer is to protect the flow through the system from disruptions, it might appear that a detailed knowledge of these disruptions-the statistical distribution curves at each step in the flow-would help (or even be a necessary element) in calculating the size of the time buffer. While this may appear to provide a rigorous methodology, it is practically useless because the required information is not available. In the practical application of DBR, we take a more pragmatic approach to establishing the size of the time buffer. Every production operation currently uses a time buffer, whether or not it is explicitly understood. By this, we mean that the production lead times used-informally or in a computerized ERP system-is many times larger than the process and setup times. All of this additional time is a time buffer. We also know that most often this currently used time buffer is much too large. This is because buffers are used to protect each step in the flow and not just the system flow as a whole and, more importantly, larger time buffers make it possible to minimize the situations in which resources simply run out of work. Since the traditional view is that a resource standing idle is a waste, lead times must be large enough to minimize idle time at each resource. In effect, one can view the current lead time as giving us an upper limit-the point where current lead time provides too large of a time buffer.

If current lead times establish one extreme for the time buffers, another extreme-a time buffer that is too small-is provided when the production lead time is close to the sum of process and setup times. In fact, in almost all production operations a production lead time that is even just three times the process time would be considered unrealistically aggressive. At each extreme end, the time buffers are actually ineffective in providing protection and promoting smooth reliable product flow. When the time buffer is too small, the cumulative disruptions that every batch of product is subject to quickly overwhelm and consume the available buffer. When the time buffer is too large, the shop floor is clogged with too much material, making it difficult to manage the flow. Each operation will have plenty of work from which to choose, and the chance that they will all be coordinated to choose the right work to promote a smooth organized flow is slim. The results are piles of inventory everywhere, long lead times, poor due date performance, and chaos on the shop floor. Between the two extremes is a range of options. Based on vast experience, we believe that Fig. 8-6 captures the essence of the effectiveness of time buffers as we increase buffer size from very small to very large. The key observation is that the curve in the region where the time buffer has a high degree of effectiveness is relatively flat. This means that there is no real benefit to complex calculations that yield precise buffer values. Being in the right ballpark is sufficient. Again, based on vast experience, a good value for the time buffer in most production environments is one-half of their current production lead time.

FIGURE 8-6 Graphical representation of the effort required to maintain a smooth flow as the buffer is increased.

The time buffer established here becomes the time element that will be used to implement the second principle of flow management (prevent overproduction). If we want to prevent production ahead of time, then we should not make the material available ahead of time. The time buffer provides the amount of the time to be used in Principle 2-a mechanism to prevent overproduction-and the rope mechanism discussed later will enable us to enforce this principle.

Whether a production operation has a true bottleneck or not, as long as there are disruptions the need for a time buffer exists. The only way to ensure that the flow at the end of the system meets promised due dates is to provide protection from disruptions using time buffers. When there is a bottleneck in the system, and this is the case when we are dealing with the full DBR system, there is a need for an additional level of protection. Any time lost at a bottleneck, by the very definition of a bottleneck, will be Throughput lost for the entire system because this lost time cannot be recovered. Hence, if one hour is lost at the bottleneck, then effectively the total system will be down for one hour and we lose the Throughput that would have been generated during this time. The downtime at a bottleneck can arise from problems at the bottleneck itself (downtime, setups, etc.) or from the same problems upstream from the bottleneck. The bottleneck can be decoupled from the disruptions upstream if we can ensure that there is always material ahead of the bottleneck. The amount of material that is sufficient to provide adequate protection depends on the nature and distribution of upstream disruptions. Note that the constraint buffer at the bottleneck is not created by adding more time into the previously established time buffer. Since the bottleneck is the true constraint to flow, material naturally accumulates at this operation/resource. All other resources have protective capacity and should be able to keep the products flowing. However, when disruptions upstream are of such a nature as to prevent the accumulation of material at the bottleneck, they threaten to create downtime at the bottleneck. This must be avoided and can be done during execution control by monitoring the amount of work at the bottleneck and taking corrective action whenever the work queue at the bottleneck is dangerously low.

It is instructive to point out here that there are other types of buffers used in the overall management of flow in a supply chain. In addition to time buffers, three other types of buffers exist: capacity buffers, stock buffers, and space buffers, with respect to production planning and control systems. The capacity buffer is defined as the protective capacity at both constraint and non-constraint resources that allow these resources to catch up when Murphy strikes. Stock buffers are defined as a "quantity of physical inventory held in the system to protect the system's throughput. Perspective: Stock buffers should not be confused with time buffers such as the constraint or shipping buffers" (Sullivan et al., 2007, 43).9 Stock buffers may be used for raw materials, WIP items (for example, at major divergent points in a V- or T-plant), and finished goods items to reduce lead time or protect against product variety. The TOCICO Dictionary defines space buffer as "Physical space immediately after the constraint that can accommodate output from the constraint when there is a stoppage downstream that would otherwise force the constraint to stop working" (Sullivan et al., 2007, 41).9 The idea is to keep the space buffer empty in the same manner as one tries to keep the constraint and shipping buffers full. BM should be used on each of these types of buffers to ensure effective operation of the constraint and high due date performance. They should also be monitored to ensure that they are not too large. Time buffers impact lead time, while stock buffers impact inventory investment.

The Rope

The final component of the DBR system is the rope-a mechanism that is used to control the flow through the system by controlling the flow at a small number of control points. The drum has created a master schedule that is consistent with the constraints of the system and is best able to satisfy customer demand. The time buffers provide the safety or insurance that the flow to the market will be reliable in spite of the impact of disruptions. The last link is to communicate effectively to the rest of the operation the actions that are necessary to support the drum and to ensure effective control of these actions.

The basic challenge is to ensure that all work centers perform the right tasks in the right sequence and at the right time. With computers becoming ubiquitous in manufacturing, it is very tempting to accomplish this objective by providing each work center with detailed schedules that are constantly updated (hopefully in real time). DBR takes a counterintuitive but far simpler approach to accomplishing this goal. The simplest and most effective way to make sure that a work center does the right job is to have only the material for the right job available. Eliminate unnecessary WIP and you eliminate opportunities for working on the wrong stuff. With this approach, the emphasis of control is shifted to strictly limiting material available at a work center to what is immediately needed. In production operations, the availability of material in the shop is controlled by the actions at the material release points-the points where raw material is released to fabrication, finished parts are released to assembly, purchased parts are released to assembly, etc.

To implement the rope, the material release points are provided with a detailed schedule that lists what materials need to be released, in what time frame, and in what sequence. If this task is managed properly, then access to unnecessary work is denied to most work centers, thereby forcing them to work on the right products. Most of the work centers that are non-constraints will simply process material when it becomes available. When a work center (a non-CCR) finds itself with more than one batch of material, what are the rules for determining the priority sequence? The real question to ask here is whether sequence really matters. In the majority of production operations, the processing time for a batch of products at any single work center is a very small fraction of the total production lead time or the total time buffer. This being the case, the difference between working on one batch before another is insignificant. It should be remembered that we are talking about very few cases where multiple batches will be available to choose from and even here, the number of batches is small. Thus, a simple rule will suffice to ensure major distortions are avoided. The priority rule can be a simple "first-in, first-out" or FIFO rule.

In simple linear flows, simply controlling the release of material will be sufficient to control the execution through the whole system. The basic principle we have followed is that we can make sure that a work center cannot work on the wrong product if material is not available. In other words, the mere fact that material is available is sufficient information to give the green light to that work center for processing. In complex flows, this basic fact is not always true. For example, at divergence points (see V-plant discussion later in this chapter), the same incoming material can be processed into different outgoing materials. It is obvious that when such a work center can be activated by material availability, we have to specify what the output products should be and how much of each product we want. While the timing of the jobs is controlled by the availability of material, workers at each divergence point (a control point) need to be provided with a detailed list of what and how much of each product to produce, as well as the priority sequence for the products.

Similar to divergence points, assembly or convergent point s may also need to be controlled. Purchased parts may be obtained in quantities larger than required for specific orders; fabrication may also have combined different orders to reduce setups at CCRs; and in T-plants (see the discussion later in the chapter), the same basic component parts can be assembled in different combinations to create different end items. The assembly departments should operate to a priority list that specifies what units should be assembled in what quantities and in what sequence.

The question often asked is, "What about at the CCR or bottleneck?" Is a detailed schedule specifying sequence and quantity of production necessary and should this be carefully monitored and controlled? If the CCR or bottleneck has sequence-dependent setups, then the setup time depends on what is currently on the resource and what product is next. A simple case is an operation that applies color. Going from a light color to a dark color requires minimal cleanup, but going from black to white will require extensive and thorough cleaning. Therefore, it is important to produce to a defined sequence and this list will have to be provided to the CCR. If this is not the case, and the process times are a small fraction of the total production lead time, then sequence even at the CCR is not that critical and no additional step beyond controlling material release is necessary. Figure 8-7 shows the schedule control points where sequence and a time frame for actions are important in controlling the flow through a factory.

Finally, there is the shipping or completion point of the batch. It is the most important schedule control point in that every batch has to meet the date when it is scheduled to be completed. Failure of a batch to meet this date or even the anticipation that a batch might fail to meet this date will trigger corrective actions as described in the section on BM.

Managing Flow with DBR-An Example

We illustrate the DBR system with a simple example10 in this section. The plant represented by the PFD in Fig. 8-8 shows a relatively simple plant with five different types of resources-each pattern in the diagram is a different type of resource (labeled R1, R2, R3, R4, R5). The number in each box in the flow diagram is the time to process one unit at that step. For example, the first step (A-1 on the bottom left of the grid used in Fig. 8-8) is performed by the R2 resource and takes 4 min for each unit. Similarly, the step corresponding to the grid point B-3 is an assembly operation performed by the R5 resource and takes 8 min per unit to assemble. To assemble a unit at B3, a unit must be completed at both A1 and C1. That assembly can then be used at the A-5 operation to make Product A or by C-5 to make Product D. The number of units of a specific resource type is indicated on the left-hand side of Fig. 8-8 and shows that there is only one resource of type R1. We also see that there are two resources of type R2. Similarly, we have two resources of types R3 and R4 and one resource of type R5. The setup time for each resource is indicated next to the resource-R1-type resource has a setup time of 15 min, R2-type resources have a setup time of 120 min, and so on. The number just below a node in the flow centric represents the number of units that are available for the resource (the WIP) at this point in time-there are 15 units available for operation E-5 performed by resource R1, etc. The demand for the three finished products is indicated at the top of the diagram and is the weekly demand for the product. For the current week, the demand for each product is as follows: 40 units of Product A, 50 units for Product D, and 40 units for Product F.

FIGURE 8-7 Schedule control points in a plant with assembly and divergence.

Based on this product structure, the demand, the material in process, and the process information for each product, we can compute the load for each type of resource. For this, we calculate the number of units that need to be processed at a given step and then multiply that by the time to process a unit. For example, at Step A-1, which feeds both Product A and Product D, the total number of units to be processed is 65 (40 units for A + 50 units for D 25 units of WIP at B-3). The total time required of the R2 resource for this production is (Number of units to be produced) (Time to produce one unit) or 65 4 min = 260 min. In a similar manner, we compute the capacity required at each step that requires an R2 resource; namely, Steps C-1, F-1, and A-5. The total load for all the steps (A-1, C-1, F-1, and A-5) comes out to 1635 min. Since there are 40 hours or 2400 minutes available in a week and since there are two R2 resources, the required time of 1635 min represents a 34 percent load [Total time required/Available time = 1635/(2 2400) = 34 percent]. The load for all resources can be calculated using this procedure. The results are shown in Table 8-1. The load as calculated here does not allow any time for setups. It is strictly the process time.

FIGURE 8-8 Product structure and resource information for sample plant. ( E. M. Goldratt used by permission, all rights reserved. Source: Modified from E. M. Goldratt (2003, 29)) From these calculations, it is clear that this operation has a CCR in the R1 resource. Of the available time, 94 percent is required just to process the units required for the week. The remaining 6 percent is available for setups, maintenance, etc. Any consumption of time (not producing product) that exceeds 6 percent will result in missed shipments. In fact, based on the information that the capacity required for processing the needed parts is 2260 at the R1 resource, we can see that only 140 min (2400 2260) are available for doing setups. Each setup requires 15 min. Therefore, we can only incur nine setups. Since there are three distinct steps where we need the R1 resource (C-5, E-5, and F-5), we conclude that we can do up to three setups for each step. To be on the safe side and allow for some fluctuations, we can choose to do two setups at each step. Effectively we will run two batches of 25 units at C-5, two batches of 25 units at E-5, and two batches of 20 units at F-5. Table 8-2 shows a schedule for the R1 resource that is constructed on this premise.

TABLE 8-1 Capacity Available and Required and Percentage Load to Satisfy Weekly Demand TABLE 8-2 Schedules for Constraint (R1 Resource) and Market for Product A Resource In this example, Product A does not require any time at the R1 resource. The market is the constraint for Product A.11 How do we manage the flow of this product? The simplest thing is to produce Product A using the customer orders. However, a single order of 40 units moving through the operation is not an example of smooth flow. To overcome the effects of large and lumpy flow in this case, we will divide the order into four batches of 10 units each and process them to be completed by the end of the week. The schedule for the R1 resource and the schedule of completions for Product A together represent the drums for this plant.

The next step is to establish the size of the time buffers. For this simple model, we select a constraint buffer of 24 hours (3 days, in this case). In real manufacturing plants, the production lead time currently being used provides the starting point. As indicated earlier, the first choice of the time buffer is to reduce this by 50 percent. For our example, we do not have this reference point. The choice of 24 hours reflects the need for a time buffer that is approximately 20 times the processing time for a unit (i.e., the process time is around 5 percent of the total production lead time). In addition, since all products have comparable routings, the time buffer is chosen to be the same for each. This means that raw material must be released 24 hours (3 days) before the expected completion by the constraint. Table 8-3 provides the rope or the release dates for the various raw materials into the process.

TABLE 8-3 The Rope (Material Release Schedule) for Sample Plant TABLE 8-4 Expected Completion Times Based on the DBR Schedule In this example, material release and the divergent point represented by operations A5 and C5 are the only schedule control points and no other information is needed for planning purposes. Table 8-2 (the drum), the choice of 24 hours as the time buffer, and Table 8-3 (the rope) provide the DBR system for this case.

An important question to answer before we commit to execution of the above DBR plan is "Does this plan help us complete the products in such a way as to meet customer expectations?" How can we identify when orders are planned to be completed. For this we have to extrapolate from the expected completion of products at the constraint (-Resource R1 as detailed in Table 8-3) by adding a reasonable estimate of time for the completion of the remaining steps (from R1 through assembly). In this simple case, we have chosen 8 hours (about one-third of the total planning lead time) for this estimate. This works in this simple case due to the relatively simple nature of the flow (minimal resource or material contention). Table 8-4 shows the time when different batches of Products A, D, and F are expected to be completed and be available for shipment. If we want to commit shipping times that can be met with very high confidence, then we should commit to hours 42 and 47 (which, in a 5-day, 8 hours per day workweek, corresponds to the Monday of week 2). In real-life situations, it is common practice to choose a slightly more conservative estimate and use one-half of the planning lead time. This means that the estimate of when a batch or an order can be completed (and hence available to ship) is equal to the completion of the last needed batch at a constraint plus one-half of the production lead time. In-transit lead time has to be added to this shipping date to determine when the order will be at the customer site.

Managing Flow-Controlling Execution and Buffer Management

The Need for Control and the Need for Corrective Actions

Using the DBR system described previously creates a plan that maximizes the system Throughput, by ensuring full utilization of the constraint while focusing on real customer demand. The plan is robust and protected from disruptions using time buffers and minimizes investment in Inventory by restricting inflow of material through the rope mechanism. This does not mean that the execution of the plan on the shop floor is automatic and that the execution does not have to be monitored carefully. It is true that in creating the time buffers, we have allowed for a certain level of disruption to the flow of a batch of material through the system. As long as the deviations actually being experienced by the batch are less than what was allowed for, we do not have a problem. However, when the actual deviation begins to exceed the allowable disruption, the ability of the batch to reach customers on time will be in jeopardy.

In these cases, not all is lost. In most manufacturing operations, there is opportunity for corrective action to be taken. The objective of these actions is to "make up" some of the time lost by the batch due to larger than anticipated disruptions. These actions include: Expediting the batch by moving it to the front of the work queue at each resource Working overtime at a resource to process this batch Processing the batch on more than one identical resource (batch splitting) Overlapping processing (carrying completed materials from one work center to the next to allow both work centers to work simultaneously) Alternate routings The use of time buffers minimizes the need for corrective actions, but it does not eliminate them. What is needed to make the DBR system deliver exceptional results in practice is a mechanism that can identify the cases where corrective action is necessary and help monitor the effectiveness of the corrective actions so that every batch can be finished on time.

Understanding Buffers: The Buffer as the Source of Information for Controlling Execution

In order to identify when a production batch is experiencing larger than "normal" disruptions, we need to go no further than understanding the time buffer in a bit more depth. When a batch of material is released one production lead time before its due date, what do we expect to happen in reality? Let us understand this by studying a sample of 100 identical batches with a production lead time of 40 days. The majority of batches experience disruptions that are within a normal but wide range. Most of the batches (about 90 percent) will reach their destination on or ahead of plan-less than or equal to 40 days. For example, some of the batches will experience far fewer than normal disruptions and these could be completed in, say, just 10 days, a time that is much shorter than the planned production lead time. Similarly, there will be a small number of batches that will experience much more than their fair share of disruptions. In the absence of corrective action, these batches will finish well past their due dates based on a 40-day planned lead time-the batches will be late. The distribution curve for the sample of 100 batches is illustrated in Fig. 8-9.

FIGURE 8-9 Graph showing the number of batches with actual lead times ranging from 10 days to 45 days where the planning lead time was 40 days with review at 35 days.

If the only point where we can identify that a batch is experiencing large disruptions is at the end of the product flow, we will have no opportunity for corrective action. We need to know that a batch in production is in trouble while there is still enough time to do something about it. What is the minimum time that will leave us enough time for corrective actions that, in the majority of cases, can help bring the batch back on track? To better understand this, let us consider a batch with a planned production lead time of 40 days that is released today. (Today is Day 1 and Due Date is Day 40.) If we simply let the normal shop floor mechanisms take place without any intervention or without even any monitoring, we expect this order to reach completion sometime between Day 10 (no major problems encountered) and Day 45 (many major problems encountered). Suppose we choose to monitor this order after 35 days have elapsed. From the statistical distribution curve shown in Fig. 8-9, approximately 70 percent of the time the order will already have been completed and the monitoring is a non-issue. However, in the remaining 30 percent of the time the monitoring will reveal the extent of the disruptions suffered; hence, the urgency of taking corrective action. In many of these cases (approximately 20 percent), the batch will be almost near completion and no action is necessary. In a small number of cases (10 percent), the batch is far behind in its progression through the shop and corrective action will be necessary. We can then initiate these corrective actions and bring this batch back on track.